paddy-data commited on
Commit
cd5ed60
·
verified ·
1 Parent(s): 75c7a0c

Remove references to internal decision records from the metadata text (2026-11-17)

Browse files
README.md CHANGED
@@ -559,7 +559,7 @@ Paddy Data: https://paddydata.ai
559
  "qw",
560
  "gripper"
561
  ],
562
- "meaning": "observation.eef_pose.<side> and observation.state's gripper of every arm in one vector: the measured TCP pose and gripper opening in a tcp action's layout, the state of an end-effector policy (ADR-0167); gripper 0 for an arm without one",
563
  "conventions": "meta/paddy/info.json conventions.eef_state"
564
  }
565
  },
@@ -780,7 +780,7 @@ Paddy Data: https://paddydata.ai
780
  "right"
781
  ],
782
  "info": {
783
- "meaning": "the arm moves at the row: TCP, an arm joint or the gripper over a window centred on the row (ADR-0187)",
784
  "rule": "meta/paddy/info.json motion"
785
  }
786
  },
@@ -794,7 +794,7 @@ Paddy Data: https://paddydata.ai
794
  "right"
795
  ],
796
  "info": {
797
- "meaning": "who drove the arm at the row (ADR-0189)",
798
  "values": {
799
  "0": "nobody",
800
  "1": "human",
 
559
  "qw",
560
  "gripper"
561
  ],
562
+ "meaning": "observation.eef_pose.<side> and observation.state's gripper of every arm in one vector: the measured TCP pose and gripper opening in a tcp action's layout, the state of an end-effector policy; gripper 0 for an arm without one",
563
  "conventions": "meta/paddy/info.json conventions.eef_state"
564
  }
565
  },
 
780
  "right"
781
  ],
782
  "info": {
783
+ "meaning": "the arm moves at the row: TCP, an arm joint or the gripper over a window centred on the row",
784
  "rule": "meta/paddy/info.json motion"
785
  }
786
  },
 
794
  "right"
795
  ],
796
  "info": {
797
+ "meaning": "who drove the arm at the row",
798
  "values": {
799
  "0": "nobody",
800
  "1": "human",
SHA256SUMS CHANGED
@@ -1,4 +1,4 @@
1
- b2d17f31acdc5982599a4854716006bb7ff61a37949800570ded680c93673711 README.md
2
  560eaa71a89c4c0fda64f33a953542215f1c7ad5c69830b5bd73b585ebce8aab data/chunk-000/file-000.parquet
3
  7a82ce2790c560b69e32e72e0e36efcf68076737cd8ed74f7d8b82e33c8233a8 data/chunk-000/file-001.parquet
4
  0f9765910c67e43481f2b7f21441c34daed9fd2444d25e1d456cc098c74cb1c9 data/chunk-000/file-002.parquet
@@ -35,16 +35,16 @@ b20ac46056f8e5dc740bb4029865152ef653b918046d977943a54694f72e6075 depth/observat
35
  fe26650697ff417c0c2ea8cd531ad018b241f2b7a129e90d9f79d85946fd3a49 depth/observation.depth_raw.camera_top/chunk-000/file-001.mkv
36
  7c0753335a263f04f3c4cb5a5b5cda6de9d942a69eb9c5b1f4ce9944b3379256 depth/observation.depth_raw.camera_top/chunk-000/file-002.mkv
37
  56aaab6ca97cea8d78f45e80abbbf8ccda3d5ac6ecec25b006a0a1ccf45533ac depth/observation.depth_raw.camera_top/chunk-000/file-003.mkv
38
- edd184fa162b15effe86629f9f1cc0493bafc690e45701031236124b5e2bfcd8 manifest.json
39
- c49c285a4b78110594b6113e86e0910d07545f731dc383f1d4ad5cf93a14524a meta/cameras.json
40
  87ec6b7368028a1ea58a55fd4f9e5f2f669804f0ad00aee624ca083625b3c8cd meta/episodes/chunk-000/file-000.parquet
41
- a6652a0da21696e9891fb88ea5e630f74e1880b69690748b7bc31cfed2a51d35 meta/info.json
42
  afc7c0cb461d05730e73c73d8928b47b91ea18351cc87d14c0ac7593f3371aeb meta/modality.json
43
- 9373c7bef737c7aa82fc50140268af035b039e2eba7a8e7f35ab2760f18d7cf3 meta/paddy/episodes/trial_001.json
44
- 5030ea6a5a6e1ce717e5c3cb8940a422ca8d54e1244227d07c6ab43eaaa1f35b meta/paddy/episodes/trial_005.json
45
- 29b436bc110a970f9521ecb32aa797d52e05dbff0f63a156f3a2b7b4d78014c2 meta/paddy/episodes/trial_007.json
46
- 0102e40ba78c2b3eb6d2c0810d1c2a32eeaf3f9f9773252acb33e47170411bc8 meta/paddy/episodes/trial_010.json
47
- 1b8a3add91a6ff2ff126d72bc5a7c66b37113cbe6ca4e3ee4249c5e962fd43ec meta/paddy/info.json
48
  649199900d8ddeb744fe30c73955ea2994396ac443848e93bcc724034198c5ff meta/paddy/keep_ranges.json
49
  5f97d29d74cf9767150b1965503ebae0dc4439429b44e002300c1836d9c28291 meta/paddy/logs/session_2026-11-14_00-40-33/trial_007/rosbag_007_camera_bottom.log
50
  651eb7233b82a8fdceeeffeff5882ef9a6be9817e6e4f16528d86ceb409ddbf0 meta/paddy/logs/session_2026-11-14_00-40-33/trial_007/rosbag_007_camera_bottom.metadata.yaml
@@ -86,8 +86,8 @@ dec07d8910239052abb81fa014a40ed1aee20ddbf8a42ec9ca53039ea9c11bd5 meta/paddy/log
86
  5cec40ac6f54b8ac61626a9370eae2d37b9af954583b8b36b2aa34ebb6e2543b meta/paddy/logs/session_2026-11-14_02-10-30/trial_005/rosbag_005_camera_top.metadata.yaml
87
  ba64023a87a0b8b519cef8b1ffb4dc6e28bff53093155265c939685b2f213bb0 meta/paddy/logs/session_2026-11-14_02-10-30/trial_005/rosbag_005_system.log
88
  011a35b790f279a1cdfa2f2949a69ce5e252ca835cdc66aae973bcc771af55ec meta/paddy/logs/session_2026-11-14_02-10-30/trial_005/rosbag_005_system.metadata.yaml
89
- 8ddf69d95926fa4ff3b4e29ec78743830d33af903f417b1ffe76b7a02349de5a meta/paddy/robot_description.urdf
90
- 71dc09e9f0326b834234f681e926a3a08148744853958b053068e96a0f6a9347 meta/paddy/stereo_calibration_camera_top.json
91
  9d83776e4a23f210d9b8c03ca0a87bbe9c233b78f4b4281e41a4d3cbeabb5541 meta/paddy/stereo_calibration_camera_top.yaml
92
  e294a408036b907a510ae0d07f1d7422c68df215154dc04a38391259dba239ec meta/stats.json
93
  94dfc34c29ce3f238b8d925b5e7c1ec840d6d844303a562bda8f033d5c65407f meta/tasks.parquet
 
1
+ 29f153d4f2870fba54c53cfa249f8dffaf6fb0f0c13eddf77211100f0f53b91d README.md
2
  560eaa71a89c4c0fda64f33a953542215f1c7ad5c69830b5bd73b585ebce8aab data/chunk-000/file-000.parquet
3
  7a82ce2790c560b69e32e72e0e36efcf68076737cd8ed74f7d8b82e33c8233a8 data/chunk-000/file-001.parquet
4
  0f9765910c67e43481f2b7f21441c34daed9fd2444d25e1d456cc098c74cb1c9 data/chunk-000/file-002.parquet
 
35
  fe26650697ff417c0c2ea8cd531ad018b241f2b7a129e90d9f79d85946fd3a49 depth/observation.depth_raw.camera_top/chunk-000/file-001.mkv
36
  7c0753335a263f04f3c4cb5a5b5cda6de9d942a69eb9c5b1f4ce9944b3379256 depth/observation.depth_raw.camera_top/chunk-000/file-002.mkv
37
  56aaab6ca97cea8d78f45e80abbbf8ccda3d5ac6ecec25b006a0a1ccf45533ac depth/observation.depth_raw.camera_top/chunk-000/file-003.mkv
38
+ 3002b97f3a3a38c4d345c518fb4871543369c6236ac3d26781078aaeb5c35a57 manifest.json
39
+ 3001440d4b585bf4aef92ddab184c1f3a3a132eae34d16edcba1eecc315ef3f6 meta/cameras.json
40
  87ec6b7368028a1ea58a55fd4f9e5f2f669804f0ad00aee624ca083625b3c8cd meta/episodes/chunk-000/file-000.parquet
41
+ 759a0ca609d7d83a3f41f119804c0184d3b7ae2754f86ddc5a2327d551024a20 meta/info.json
42
  afc7c0cb461d05730e73c73d8928b47b91ea18351cc87d14c0ac7593f3371aeb meta/modality.json
43
+ 3ebe756c799078a85cd6c275898adb74dcf0fa5e8c80e9028455fc3a8b651169 meta/paddy/episodes/trial_001.json
44
+ e11c58fcfdffe3d439063b10490bc9154a52427c1fc5ac29ab123cee07231b5c meta/paddy/episodes/trial_005.json
45
+ d41af2a0c5da5c101ecbdebf356cf3349135dd02f5f80a5e84cc125516124895 meta/paddy/episodes/trial_007.json
46
+ 9caf3e6a88829fa32d14e304aa964ea870a67f137fc9a8b5b0e5aafeb5abb80a meta/paddy/episodes/trial_010.json
47
+ 9fb7f925ff7a8781ea16710f96293f346f5096ba09d6572b97c85f5f525b9345 meta/paddy/info.json
48
  649199900d8ddeb744fe30c73955ea2994396ac443848e93bcc724034198c5ff meta/paddy/keep_ranges.json
49
  5f97d29d74cf9767150b1965503ebae0dc4439429b44e002300c1836d9c28291 meta/paddy/logs/session_2026-11-14_00-40-33/trial_007/rosbag_007_camera_bottom.log
50
  651eb7233b82a8fdceeeffeff5882ef9a6be9817e6e4f16528d86ceb409ddbf0 meta/paddy/logs/session_2026-11-14_00-40-33/trial_007/rosbag_007_camera_bottom.metadata.yaml
 
86
  5cec40ac6f54b8ac61626a9370eae2d37b9af954583b8b36b2aa34ebb6e2543b meta/paddy/logs/session_2026-11-14_02-10-30/trial_005/rosbag_005_camera_top.metadata.yaml
87
  ba64023a87a0b8b519cef8b1ffb4dc6e28bff53093155265c939685b2f213bb0 meta/paddy/logs/session_2026-11-14_02-10-30/trial_005/rosbag_005_system.log
88
  011a35b790f279a1cdfa2f2949a69ce5e252ca835cdc66aae973bcc771af55ec meta/paddy/logs/session_2026-11-14_02-10-30/trial_005/rosbag_005_system.metadata.yaml
89
+ 29c567e3f0d6c8c4bf52e66b19873b0fe75d21af0d69064c9adc4464171a93eb meta/paddy/robot_description.urdf
90
+ d068fce95c3feb17d2fb1912556efe2d8fa4940a4ce3a9d1f4791dbe10cd20ef meta/paddy/stereo_calibration_camera_top.json
91
  9d83776e4a23f210d9b8c03ca0a87bbe9c233b78f4b4281e41a4d3cbeabb5541 meta/paddy/stereo_calibration_camera_top.yaml
92
  e294a408036b907a510ae0d07f1d7422c68df215154dc04a38391259dba239ec meta/stats.json
93
  94dfc34c29ce3f238b8d925b5e7c1ec840d6d844303a562bda8f033d5c65407f meta/tasks.parquet
manifest.json CHANGED
@@ -57,7 +57,7 @@
57
  "camera_right": "wrist"
58
  },
59
  "files": 132,
60
- "bytes": 19379759798
61
  },
62
  "known_issues": [],
63
  "episodes": [
@@ -117,8 +117,8 @@
117
  "files": [
118
  {
119
  "path": "README.md",
120
- "bytes": 47028,
121
- "sha256": "b2d17f31acdc5982599a4854716006bb7ff61a37949800570ded680c93673711"
122
  },
123
  {
124
  "path": "data/chunk-000/file-000.parquet",
@@ -302,8 +302,8 @@
302
  },
303
  {
304
  "path": "meta/cameras.json",
305
- "bytes": 46749,
306
- "sha256": "c49c285a4b78110594b6113e86e0910d07545f731dc383f1d4ad5cf93a14524a"
307
  },
308
  {
309
  "path": "meta/episodes/chunk-000/file-000.parquet",
@@ -312,8 +312,8 @@
312
  },
313
  {
314
  "path": "meta/info.json",
315
- "bytes": 24137,
316
- "sha256": "a6652a0da21696e9891fb88ea5e630f74e1880b69690748b7bc31cfed2a51d35"
317
  },
318
  {
319
  "path": "meta/modality.json",
@@ -322,28 +322,28 @@
322
  },
323
  {
324
  "path": "meta/paddy/episodes/trial_001.json",
325
- "bytes": 163699,
326
- "sha256": "9373c7bef737c7aa82fc50140268af035b039e2eba7a8e7f35ab2760f18d7cf3"
327
  },
328
  {
329
  "path": "meta/paddy/episodes/trial_005.json",
330
- "bytes": 148906,
331
- "sha256": "5030ea6a5a6e1ce717e5c3cb8940a422ca8d54e1244227d07c6ab43eaaa1f35b"
332
  },
333
  {
334
  "path": "meta/paddy/episodes/trial_007.json",
335
- "bytes": 149785,
336
- "sha256": "29b436bc110a970f9521ecb32aa797d52e05dbff0f63a156f3a2b7b4d78014c2"
337
  },
338
  {
339
  "path": "meta/paddy/episodes/trial_010.json",
340
- "bytes": 146949,
341
- "sha256": "0102e40ba78c2b3eb6d2c0810d1c2a32eeaf3f9f9773252acb33e47170411bc8"
342
  },
343
  {
344
  "path": "meta/paddy/info.json",
345
- "bytes": 112626,
346
- "sha256": "1b8a3add91a6ff2ff126d72bc5a7c66b37113cbe6ca4e3ee4249c5e962fd43ec"
347
  },
348
  {
349
  "path": "meta/paddy/keep_ranges.json",
@@ -552,13 +552,13 @@
552
  },
553
  {
554
  "path": "meta/paddy/robot_description.urdf",
555
- "bytes": 75247,
556
- "sha256": "8ddf69d95926fa4ff3b4e29ec78743830d33af903f417b1ffe76b7a02349de5a"
557
  },
558
  {
559
  "path": "meta/paddy/stereo_calibration_camera_top.json",
560
- "bytes": 20206,
561
- "sha256": "71dc09e9f0326b834234f681e926a3a08148744853958b053068e96a0f6a9347"
562
  },
563
  {
564
  "path": "meta/paddy/stereo_calibration_camera_top.yaml",
 
57
  "camera_right": "wrist"
58
  },
59
  "files": 132,
60
+ "bytes": 19379758481
61
  },
62
  "known_issues": [],
63
  "episodes": [
 
117
  "files": [
118
  {
119
  "path": "README.md",
120
+ "bytes": 46995,
121
+ "sha256": "29f153d4f2870fba54c53cfa249f8dffaf6fb0f0c13eddf77211100f0f53b91d"
122
  },
123
  {
124
  "path": "data/chunk-000/file-000.parquet",
 
302
  },
303
  {
304
  "path": "meta/cameras.json",
305
+ "bytes": 46717,
306
+ "sha256": "3001440d4b585bf4aef92ddab184c1f3a3a132eae34d16edcba1eecc315ef3f6"
307
  },
308
  {
309
  "path": "meta/episodes/chunk-000/file-000.parquet",
 
312
  },
313
  {
314
  "path": "meta/info.json",
315
+ "bytes": 24104,
316
+ "sha256": "759a0ca609d7d83a3f41f119804c0184d3b7ae2754f86ddc5a2327d551024a20"
317
  },
318
  {
319
  "path": "meta/modality.json",
 
322
  },
323
  {
324
  "path": "meta/paddy/episodes/trial_001.json",
325
+ "bytes": 163677,
326
+ "sha256": "3ebe756c799078a85cd6c275898adb74dcf0fa5e8c80e9028455fc3a8b651169"
327
  },
328
  {
329
  "path": "meta/paddy/episodes/trial_005.json",
330
+ "bytes": 148884,
331
+ "sha256": "e11c58fcfdffe3d439063b10490bc9154a52427c1fc5ac29ab123cee07231b5c"
332
  },
333
  {
334
  "path": "meta/paddy/episodes/trial_007.json",
335
+ "bytes": 149763,
336
+ "sha256": "d41af2a0c5da5c101ecbdebf356cf3349135dd02f5f80a5e84cc125516124895"
337
  },
338
  {
339
  "path": "meta/paddy/episodes/trial_010.json",
340
+ "bytes": 146927,
341
+ "sha256": "9caf3e6a88829fa32d14e304aa964ea870a67f137fc9a8b5b0e5aafeb5abb80a"
342
  },
343
  {
344
  "path": "meta/paddy/info.json",
345
+ "bytes": 111561,
346
+ "sha256": "9fb7f925ff7a8781ea16710f96293f346f5096ba09d6572b97c85f5f525b9345"
347
  },
348
  {
349
  "path": "meta/paddy/keep_ranges.json",
 
552
  },
553
  {
554
  "path": "meta/paddy/robot_description.urdf",
555
+ "bytes": 75192,
556
+ "sha256": "29c567e3f0d6c8c4bf52e66b19873b0fe75d21af0d69064c9adc4464171a93eb"
557
  },
558
  {
559
  "path": "meta/paddy/stereo_calibration_camera_top.json",
560
+ "bytes": 20195,
561
+ "sha256": "d068fce95c3feb17d2fb1912556efe2d8fa4940a4ce3a9d1f4791dbe10cd20ef"
562
  },
563
  {
564
  "path": "meta/paddy/stereo_calibration_camera_top.yaml",
meta/cameras.json CHANGED
@@ -467,7 +467,7 @@
467
  "calibrated_at": "2026-11-06",
468
  "error_px": 2.39,
469
  "samples": 122,
470
- "notes": "Fitted at 848x480 (RMS 1.06 px there, 2.39 px scaled to 1920x1080), carried to 1280x720 (ADR-0206) and on 2026-11-13 to 1920x1080 with the factory K at each size; ChArUco table board 9x7, 60 mm, held in 122 still views (4619 corners, 0.27 to 0.78 m, tilt median 40 deg); capture top_intrinsics_20261106_180743; RMS 1.20 px with D = 0, 1.06 fitted; corners beyond 0.82 of the half-diagonal extrapolated (up to 1.9 px)",
471
  "from": "rig_snapshot"
472
  },
473
  "mount": {
@@ -548,7 +548,7 @@
548
  "error_mm": 0.43,
549
  "error_px": 0.55,
550
  "samples": 31,
551
- "notes": "ChArUco table board 9x7, 60 mm squares as measured; capture calib_20261112_173947; intrinsics measured 2026-11-06 (ADR-0181)",
552
  "from": "rig_snapshot"
553
  },
554
  "depth": {
@@ -844,7 +844,7 @@
844
  "emitter": false,
845
  "image_rotation_deg": 0,
846
  "pairing": "same header stamp",
847
- "source": "measured stereo calibration (meta/paddy/stereo_calibration_camera_top.json, ADR-0228)",
848
  "calibration_source": "measured",
849
  "calibration": {
850
  "source": "measured",
 
467
  "calibrated_at": "2026-11-06",
468
  "error_px": 2.39,
469
  "samples": 122,
470
+ "notes": "Fitted at 848x480 (RMS 1.06 px there, 2.39 px scaled to 1920x1080), carried to 1280x720 and on 2026-11-13 to 1920x1080 with the factory K at each size; ChArUco table board 9x7, 60 mm, held in 122 still views (4619 corners, 0.27 to 0.78 m, tilt median 40 deg); capture top_intrinsics_20261106_180743; RMS 1.20 px with D = 0, 1.06 fitted; corners beyond 0.82 of the half-diagonal extrapolated (up to 1.9 px)",
471
  "from": "rig_snapshot"
472
  },
473
  "mount": {
 
548
  "error_mm": 0.43,
549
  "error_px": 0.55,
550
  "samples": 31,
551
+ "notes": "ChArUco table board 9x7, 60 mm squares as measured; capture calib_20261112_173947; intrinsics measured 2026-11-06",
552
  "from": "rig_snapshot"
553
  },
554
  "depth": {
 
844
  "emitter": false,
845
  "image_rotation_deg": 0,
846
  "pairing": "same header stamp",
847
+ "source": "measured stereo calibration (meta/paddy/stereo_calibration_camera_top.json)",
848
  "calibration_source": "measured",
849
  "calibration": {
850
  "source": "measured",
meta/info.json CHANGED
@@ -252,7 +252,7 @@
252
  "qw",
253
  "gripper"
254
  ],
255
- "meaning": "observation.eef_pose.<side> and observation.state's gripper of every arm in one vector: the measured TCP pose and gripper opening in a tcp action's layout, the state of an end-effector policy (ADR-0167); gripper 0 for an arm without one",
256
  "conventions": "meta/paddy/info.json conventions.eef_state"
257
  }
258
  },
@@ -473,7 +473,7 @@
473
  "right"
474
  ],
475
  "info": {
476
- "meaning": "the arm moves at the row: TCP, an arm joint or the gripper over a window centred on the row (ADR-0187)",
477
  "rule": "meta/paddy/info.json motion"
478
  }
479
  },
@@ -487,7 +487,7 @@
487
  "right"
488
  ],
489
  "info": {
490
- "meaning": "who drove the arm at the row (ADR-0189)",
491
  "values": {
492
  "0": "nobody",
493
  "1": "human",
 
252
  "qw",
253
  "gripper"
254
  ],
255
+ "meaning": "observation.eef_pose.<side> and observation.state's gripper of every arm in one vector: the measured TCP pose and gripper opening in a tcp action's layout, the state of an end-effector policy; gripper 0 for an arm without one",
256
  "conventions": "meta/paddy/info.json conventions.eef_state"
257
  }
258
  },
 
473
  "right"
474
  ],
475
  "info": {
476
+ "meaning": "the arm moves at the row: TCP, an arm joint or the gripper over a window centred on the row",
477
  "rule": "meta/paddy/info.json motion"
478
  }
479
  },
 
487
  "right"
488
  ],
489
  "info": {
490
+ "meaning": "who drove the arm at the row",
491
  "values": {
492
  "0": "nobody",
493
  "1": "human",
meta/paddy/episodes/trial_001.json CHANGED
@@ -4235,7 +4235,7 @@
4235
  "error_px": 0.55,
4236
  "frame": "Right_link_base",
4237
  "method": "paddy_charuco_robot_world",
4238
- "notes": "ChArUco table board 9x7, 60 mm squares as measured; capture calib_20261112_173947; intrinsics measured 2026-11-06 (ADR-0181)",
4239
  "rpy_deg": [
4240
  -59.04,
4241
  -88.19,
@@ -4276,7 +4276,7 @@
4276
  "fy": 1361.3901,
4277
  "height": 1080,
4278
  "method": "opencv_calibrate_factory_k",
4279
- "notes": "Fitted at 848x480 (RMS 1.06 px there, 2.39 px scaled to 1920x1080), carried to 1280x720 (ADR-0206) and on 2026-11-13 to 1920x1080 with the factory K at each size; ChArUco table board 9x7, 60 mm, held in 122 still views (4619 corners, 0.27 to 0.78 m, tilt median 40 deg); capture top_intrinsics_20261106_180743; RMS 1.20 px with D = 0, 1.06 fitted; corners beyond 0.82 of the half-diagonal extrapolated (up to 1.9 px)",
4280
  "samples": 122,
4281
  "width": 1920
4282
  },
 
4235
  "error_px": 0.55,
4236
  "frame": "Right_link_base",
4237
  "method": "paddy_charuco_robot_world",
4238
+ "notes": "ChArUco table board 9x7, 60 mm squares as measured; capture calib_20261112_173947; intrinsics measured 2026-11-06",
4239
  "rpy_deg": [
4240
  -59.04,
4241
  -88.19,
 
4276
  "fy": 1361.3901,
4277
  "height": 1080,
4278
  "method": "opencv_calibrate_factory_k",
4279
+ "notes": "Fitted at 848x480 (RMS 1.06 px there, 2.39 px scaled to 1920x1080), carried to 1280x720 and on 2026-11-13 to 1920x1080 with the factory K at each size; ChArUco table board 9x7, 60 mm, held in 122 still views (4619 corners, 0.27 to 0.78 m, tilt median 40 deg); capture top_intrinsics_20261106_180743; RMS 1.20 px with D = 0, 1.06 fitted; corners beyond 0.82 of the half-diagonal extrapolated (up to 1.9 px)",
4280
  "samples": 122,
4281
  "width": 1920
4282
  },
meta/paddy/episodes/trial_005.json CHANGED
@@ -3733,7 +3733,7 @@
3733
  "error_px": 0.55,
3734
  "frame": "Right_link_base",
3735
  "method": "paddy_charuco_robot_world",
3736
- "notes": "ChArUco table board 9x7, 60 mm squares as measured; capture calib_20261112_173947; intrinsics measured 2026-11-06 (ADR-0181)",
3737
  "rpy_deg": [
3738
  -59.04,
3739
  -88.19,
@@ -3774,7 +3774,7 @@
3774
  "fy": 1361.3901,
3775
  "height": 1080,
3776
  "method": "opencv_calibrate_factory_k",
3777
- "notes": "Fitted at 848x480 (RMS 1.06 px there, 2.39 px scaled to 1920x1080), carried to 1280x720 (ADR-0206) and on 2026-11-13 to 1920x1080 with the factory K at each size; ChArUco table board 9x7, 60 mm, held in 122 still views (4619 corners, 0.27 to 0.78 m, tilt median 40 deg); capture top_intrinsics_20261106_180743; RMS 1.20 px with D = 0, 1.06 fitted; corners beyond 0.82 of the half-diagonal extrapolated (up to 1.9 px)",
3778
  "samples": 122,
3779
  "width": 1920
3780
  },
 
3733
  "error_px": 0.55,
3734
  "frame": "Right_link_base",
3735
  "method": "paddy_charuco_robot_world",
3736
+ "notes": "ChArUco table board 9x7, 60 mm squares as measured; capture calib_20261112_173947; intrinsics measured 2026-11-06",
3737
  "rpy_deg": [
3738
  -59.04,
3739
  -88.19,
 
3774
  "fy": 1361.3901,
3775
  "height": 1080,
3776
  "method": "opencv_calibrate_factory_k",
3777
+ "notes": "Fitted at 848x480 (RMS 1.06 px there, 2.39 px scaled to 1920x1080), carried to 1280x720 and on 2026-11-13 to 1920x1080 with the factory K at each size; ChArUco table board 9x7, 60 mm, held in 122 still views (4619 corners, 0.27 to 0.78 m, tilt median 40 deg); capture top_intrinsics_20261106_180743; RMS 1.20 px with D = 0, 1.06 fitted; corners beyond 0.82 of the half-diagonal extrapolated (up to 1.9 px)",
3778
  "samples": 122,
3779
  "width": 1920
3780
  },
meta/paddy/episodes/trial_007.json CHANGED
@@ -3744,7 +3744,7 @@
3744
  "error_px": 0.55,
3745
  "frame": "Right_link_base",
3746
  "method": "paddy_charuco_robot_world",
3747
- "notes": "ChArUco table board 9x7, 60 mm squares as measured; capture calib_20261112_173947; intrinsics measured 2026-11-06 (ADR-0181)",
3748
  "rpy_deg": [
3749
  -59.04,
3750
  -88.19,
@@ -3785,7 +3785,7 @@
3785
  "fy": 1361.3901,
3786
  "height": 1080,
3787
  "method": "opencv_calibrate_factory_k",
3788
- "notes": "Fitted at 848x480 (RMS 1.06 px there, 2.39 px scaled to 1920x1080), carried to 1280x720 (ADR-0206) and on 2026-11-13 to 1920x1080 with the factory K at each size; ChArUco table board 9x7, 60 mm, held in 122 still views (4619 corners, 0.27 to 0.78 m, tilt median 40 deg); capture top_intrinsics_20261106_180743; RMS 1.20 px with D = 0, 1.06 fitted; corners beyond 0.82 of the half-diagonal extrapolated (up to 1.9 px)",
3789
  "samples": 122,
3790
  "width": 1920
3791
  },
 
3744
  "error_px": 0.55,
3745
  "frame": "Right_link_base",
3746
  "method": "paddy_charuco_robot_world",
3747
+ "notes": "ChArUco table board 9x7, 60 mm squares as measured; capture calib_20261112_173947; intrinsics measured 2026-11-06",
3748
  "rpy_deg": [
3749
  -59.04,
3750
  -88.19,
 
3785
  "fy": 1361.3901,
3786
  "height": 1080,
3787
  "method": "opencv_calibrate_factory_k",
3788
+ "notes": "Fitted at 848x480 (RMS 1.06 px there, 2.39 px scaled to 1920x1080), carried to 1280x720 and on 2026-11-13 to 1920x1080 with the factory K at each size; ChArUco table board 9x7, 60 mm, held in 122 still views (4619 corners, 0.27 to 0.78 m, tilt median 40 deg); capture top_intrinsics_20261106_180743; RMS 1.20 px with D = 0, 1.06 fitted; corners beyond 0.82 of the half-diagonal extrapolated (up to 1.9 px)",
3789
  "samples": 122,
3790
  "width": 1920
3791
  },
meta/paddy/episodes/trial_010.json CHANGED
@@ -3655,7 +3655,7 @@
3655
  "error_px": 0.55,
3656
  "frame": "Right_link_base",
3657
  "method": "paddy_charuco_robot_world",
3658
- "notes": "ChArUco table board 9x7, 60 mm squares as measured; capture calib_20261112_173947; intrinsics measured 2026-11-06 (ADR-0181)",
3659
  "rpy_deg": [
3660
  -59.04,
3661
  -88.19,
@@ -3696,7 +3696,7 @@
3696
  "fy": 1361.3901,
3697
  "height": 1080,
3698
  "method": "opencv_calibrate_factory_k",
3699
- "notes": "Fitted at 848x480 (RMS 1.06 px there, 2.39 px scaled to 1920x1080), carried to 1280x720 (ADR-0206) and on 2026-11-13 to 1920x1080 with the factory K at each size; ChArUco table board 9x7, 60 mm, held in 122 still views (4619 corners, 0.27 to 0.78 m, tilt median 40 deg); capture top_intrinsics_20261106_180743; RMS 1.20 px with D = 0, 1.06 fitted; corners beyond 0.82 of the half-diagonal extrapolated (up to 1.9 px)",
3700
  "samples": 122,
3701
  "width": 1920
3702
  },
 
3655
  "error_px": 0.55,
3656
  "frame": "Right_link_base",
3657
  "method": "paddy_charuco_robot_world",
3658
+ "notes": "ChArUco table board 9x7, 60 mm squares as measured; capture calib_20261112_173947; intrinsics measured 2026-11-06",
3659
  "rpy_deg": [
3660
  -59.04,
3661
  -88.19,
 
3696
  "fy": 1361.3901,
3697
  "height": 1080,
3698
  "method": "opencv_calibrate_factory_k",
3699
+ "notes": "Fitted at 848x480 (RMS 1.06 px there, 2.39 px scaled to 1920x1080), carried to 1280x720 and on 2026-11-13 to 1920x1080 with the factory K at each size; ChArUco table board 9x7, 60 mm, held in 122 still views (4619 corners, 0.27 to 0.78 m, tilt median 40 deg); capture top_intrinsics_20261106_180743; RMS 1.20 px with D = 0, 1.06 fitted; corners beyond 0.82 of the half-diagonal extrapolated (up to 1.9 px)",
3700
  "samples": 122,
3701
  "width": 1920
3702
  },
meta/paddy/info.json CHANGED
@@ -250,7 +250,7 @@
250
  "action": {
251
  "feature": "action",
252
  "space": "joints",
253
- "meaning": "per arm the joint command, zero-order held, then the gripper command: the quantity a policy that drives the arms through target_joints outputs (ADR-0161). The TCP target the teleop controller published is paddy.target_pose.<side>; the pose the follower sent is paddy.command_sent.<side>",
254
  "joints": "absolute positions of the URDF joints joint1..jointN, base to wrist, in observation.state's order; not a delta, not a TCP pose",
255
  "joint_unit": "rad",
256
  "joints_source": {
@@ -259,7 +259,7 @@
259
  },
260
  "joints_sources": {
261
  "command_sent": "the joint angles the follower sent (command_sent joints_rad: a set_servo_angle call's target or, in servo mode, a set_servo_angle_j setpoint; the xArm7 joint path, GELLO and policy arms), held with action_hold_s; timing.command_methods says which",
262
- "target_joints": "a leader arm's recorded joint targets (target_joints, GELLO, ADR-0097), held with action_hold_s",
263
  "next_state": "the next row's measured joints (the last row its own): the usual stand-in for a position-controlled arm's joint command when none was recorded (a mode-7 arm, older bags)"
264
  },
265
  "joints_source_per_episode": "conversion.action_source.<side> of meta/paddy/episodes/<trial_id>.json",
@@ -320,10 +320,10 @@
320
  }
321
  },
322
  "already_scaled": true,
323
- "scaling_note": "action is in robot units after the controller's motion scaling and rate limit (ADR-0046); paddy.scaling is the human-input gain in force (informational, not a policy input)"
324
  },
325
  "timing": {
326
- "row_time": "t_k = t0 + k / fps (t0 = the latest first stamp over the arms' joint states and the cameras' color frames, ADR-0173; paused spans removed); timestamp = k / fps; paddy.utc_s and the sidecar's conversion.t_start_s give the absolute time",
327
  "rule": "row k holds the last joint command stamped at or before t_k (zero-order hold, no shift): the command issued at t_k; with joints_source next_state the measured joints of row k+1, the position the arm reached next",
328
  "stamp_meaning": "command_sent: header.stamp of the command_sent message, when the follower's motion call (set_servo_angle, or set_servo_angle_j in servo mode) returned; target_joints: header.stamp of the leader's target; next_state: the next row's slot time",
329
  "execution": "the follower validates each target and commands the arm from it; how depends on the arm's control mode, named per command by command_sent.method (per arm command_methods, per episode the sidecar's conversion.streams.command_sent.<side>.methods) and described in execution_by_method. The issue time of each command is header.stamp of the command_sent stream",
@@ -336,7 +336,7 @@
336
  ]
337
  },
338
  "execution_by_method": {
339
- "set_servo_angle_j": "servo mode (xArm mode 1, ADR-0177): the follower takes each accepted target as joints (its own solution of a TCP target, or a leader's joint target), steps a joint setpoint toward the newest one on a fixed tick (rates.control.servo_rate_hz), each joint inside the stream's speed and acceleration caps, and sends every setpoint with set_servo_angle_j; the control box executes the last setpoint it received and plans nothing. command_sent has one entry per setpoint: joints_rad is the setpoint, the pose and target_stamp are those of the target it steps toward"
340
  },
341
  "action_hold_s": 0.2,
342
  "valid_mask": "paddy.action_valid",
@@ -382,12 +382,12 @@
382
  "left": "urdf_fk",
383
  "right": "urdf_fk"
384
  },
385
- "sources": "urdf_fk: the recorded URDF's base_T_tcp at the recorded joint states (100 Hz), the kinematics the wrist cameras' paddy.camera_pose and the teleop targets use (ADR-0182); reported: the pose the arm's control box reports (about 50 Hz, its own kinematics, which can differ from the URDF's by several mm: sidecar conversion.kinematics.<side>.tcp_vs_urdf_mm); mixed: episodes differ, see each sidecar's conversion.tcp_pose.<side>",
386
- "stamps": "urdf_fk: the joint states' stamps. reported: the control box reports the pose ahead of the arm (ADR-0175); each sample is restamped by that lead, measured per episode against the URDF's TCP at the recorded joints, else the rig profile's actual_pose_lead_ms: sidecar conversion.eef_timing.<side>"
387
  },
388
  "eef_state": {
389
  "feature": "observation.eef_state",
390
- "meaning": "every arm's measured TCP pose (observation.eef_pose.<side>, sign-continuous) and gripper opening (observation.state's) in one vector, the measured counterpart of a tcp action: the state of an end-effector policy (ADR-0167)",
391
  "layout_per_arm": [
392
  "x",
393
  "y",
@@ -406,29 +406,29 @@
406
  },
407
  "kinematics": {
408
  "sidecar": "conversion.kinematics.<side> of meta/paddy/episodes/<trial_id>.json",
409
- "tcp_vs_urdf": "tcp_vs_urdf_mm and tcp_vs_urdf_deg (median, max): the URDF's base_T_tcp at the recorded joints against the recorded TCP pose (ADR-0078)",
410
- "dh": "dh: the arm control box's own DH table (metadata.json controller_settings.<side>.dh_params: 28 values, four per joint, mm and rad) as identified against the recorded TCP pose: layout (the name of each of the four values per joint, in the table's order), convention (modified = Craig, or standard), tcp_vs_dh_mm and tcp_vs_dh_deg (median, max), samples, fit, and dh_vs_urdf_mm (the table's flange against the URDF's at the same joints); with fit true, forward kinematics of the table read that way is the arm's calibrated kinematics. {note} when the trial has no table (ADR-0131)"
411
  },
412
  "cameras": {
413
  "file": "meta/cameras.json",
414
- "role": "where the camera is mounted, as the rig profile declares it (ADR-0184): overhead (fixed, looking down at the workspace), scene (fixed, another viewpoint) or wrist (moves with an arm; which arm is mount.side); null = not declared",
415
  "primary": "the file's top-level primary names the rig's one primary camera (the view to look at first), and that camera's entry has primary: true; null = not declared",
416
  "base_T_cam": "4x4 pose of the camera's color optical frame (z forward, x right, y down) in each arm base frame, metres; extrinsics.<base>.source says where it came from",
417
  "sources": {
418
  "tf_static": "the nominal pose: the rig's URDF and the camera mounts in the rig profile (CAD or tape), as published on /tf_static",
419
  "tf_snapshot": "the nominal pose from the first /tf snapshot at episode start (valid: episode_start)",
420
- "urdf_fk": "the nominal pose of a wrist camera by forward kinematics of the recorded URDF and joint states at t_s, the episode's first slot (valid: episode_start, ADR-0078)",
421
- "measured": "the rig profile's measured calibration (ADR-0132): calibration.frame_T_cam chained to the base through via (identity, tf_static, urdf, urdf_fk at t_s for a wrist camera, or tf_snapshot); the nominal entry is kept as nominal {base_T_cam, source}"
422
  },
423
  "measured_vs_nominal": "measured_vs_nominal_mm / _deg: distance of the origins and angle of the rotation between the measured and the nominal base_T_cam at the episode's first slot",
424
- "intrinsics": "K (row-major 3x3) and D (plumb_bob k1 k2 p1 p2 k3) describe the raw color image as recorded (not undistorted). intrinsics.source camera_info: the camera's own CameraInfo (a RealSense's factory values; a D435 color stream reports D all zeros). measured: the rig profile's measured intrinsics (ADR-0181: method, calibrated_at, error_px, samples, notes, from), the CameraInfo values kept as factory_intrinsics {K, D, distortion_model}. The pose of a fixed camera measured after its intrinsics were is solved with them",
425
  "calibration": "per camera: the measured calibration (frame, xyz_mm, rpy_deg URDF fixed-axis, frame_T_cam, method, calibrated_at, error_mm, error_px, samples, notes, from: rig_snapshot | rig_option), null when the rig profile has none. TF on the rig stays nominal",
426
- "image_rotation": "image_rotation_deg per camera (ADR-0194): 0, or 180 for a camera mounted upside down whose pictures the dataset turns. Turned, the color and depth frames are the recorded ones turned by 180 degrees and every value that describes them is of the turned image: K (cx' = W - 1 - cx, cy' = H - 1 - cy), D (p1 and p2 negated), base_T_cam, mount link_T_cam and tcp_T_cam, calibration.frame_T_cam, the depth K, D and color_T_depth, and paddy.camera_pose.<camera>: the camera's own optical frame (frame_id) turned about its z axis, X_T_turned = X_T_cam @ diag(-1, -1, 1, 1). calibration.xyz_mm and rpy_deg stay the rig profile's record of the camera's own optical frame. A policy trained on it needs the live frames turned the same way (paddy's policy server does it from this file)",
427
- "per_row_pose": "paddy.camera_pose.<camera> of a wrist camera; source measured_fk (forward kinematics to the calibration frame times its measured pose), urdf_fk or eef; time row (at the joint sample observation.state holds, so it matches observation.state through the URDF; the image is paddy.frame_age.<camera> older) or frame (at the shown frame's stamp), ADR-0179"
428
  },
429
  "subtasks": {
430
  "feature": "paddy.subtask_index",
431
- "meaning": "the operator-marked subtask open at the row (metadata.json subtasks, ADR-0133): its index, -1 where none is open. Spans are half-open [start, end) in header.stamp POSIX seconds; paused spans are out of the grid, so a subtask with a pause inside has no rows for the paused time",
432
  "names": "the episode sidecar's conversion.subtasks: per index the plain-language instruction, the outcome (success, failure or unknown), start_s, end_s (null: still open at the end of the recording) and the rows [start_row, end_row) it covers (null when none)",
433
  "not_a_policy_input": true
434
  },
@@ -493,12 +493,12 @@
493
  "frames": {
494
  "right": "Right_link_eef"
495
  },
496
- "frame_note": "the flange (URDF link_eef), which the six-axis sensor is mounted on (ADR-0080). Checked on the rig (validation 1.26, 2026-11-07): a press along each flange axis reads on that axis with the sign of the force applied to the tool (a push toward flange +x reads +fx); a load fit put the sensor axes within 1.4 deg of the flange's",
497
  "reading": "observation.force_torque.<side>: the control box's reading after its load and offset compensation (/paddy/<side>_arm/ft_wrench, SDK ft_ext_force): near zero, up to a small offset, with nothing touching the tool when the payload set on the control box matches the tool. paddy.force_torque_raw.<side>: the unprocessed reading (/paddy/<side>_arm/ft_wrench_raw, SDK ft_raw_force), so the compensation can be redone",
498
  "compensation": "the payload the control box compensated for is in each episode sidecar's metadata.controller_settings.<side>.tcp_load_kg and tcp_load_cog_mm, read at the trial's start",
499
- "timing": "row k holds the last reading stamped at or before t_k (zero-order hold, fresh within hold_max_age_s); the stamp is when the follower's host read the control box's real-time report packet (ADR-0081)",
500
  "missing": "rows without a fresh reading hold zeros and paddy.force_torque_valid false; they are not dropped. An arm whose rig declares the sensor but recorded no readings has the columns, all zeros and invalid",
501
- "rate_hz_nominal": "about 100 (the real-time report); 5 when the follower fell back to the rich report (ADR-0081)",
502
  "measured_median_hz": {
503
  "right": 99.5
504
  },
@@ -567,7 +567,7 @@
567
  "action": {
568
  "feature": "action",
569
  "space": "joints",
570
- "meaning": "per arm the joint command, zero-order held, then the gripper command: the quantity a policy that drives the arms through target_joints outputs (ADR-0161). The TCP target the teleop controller published is paddy.target_pose.<side>; the pose the follower sent is paddy.command_sent.<side>",
571
  "joints": "absolute positions of the URDF joints joint1..jointN, base to wrist, in observation.state's order; not a delta, not a TCP pose",
572
  "joint_unit": "rad",
573
  "joints_source": {
@@ -576,7 +576,7 @@
576
  },
577
  "joints_sources": {
578
  "command_sent": "the joint angles the follower sent (command_sent joints_rad: a set_servo_angle call's target or, in servo mode, a set_servo_angle_j setpoint; the xArm7 joint path, GELLO and policy arms), held with action_hold_s; timing.command_methods says which",
579
- "target_joints": "a leader arm's recorded joint targets (target_joints, GELLO, ADR-0097), held with action_hold_s",
580
  "next_state": "the next row's measured joints (the last row its own): the usual stand-in for a position-controlled arm's joint command when none was recorded (a mode-7 arm, older bags)"
581
  },
582
  "joints_source_per_episode": "conversion.action_source.<side> of meta/paddy/episodes/<trial_id>.json",
@@ -637,10 +637,10 @@
637
  }
638
  },
639
  "already_scaled": true,
640
- "scaling_note": "action is in robot units after the controller's motion scaling and rate limit (ADR-0046); paddy.scaling is the human-input gain in force (informational, not a policy input)"
641
  },
642
  "timing": {
643
- "row_time": "t_k = t0 + k / fps (t0 = the latest first stamp over the arms' joint states and the cameras' color frames, ADR-0173; paused spans removed); timestamp = k / fps; paddy.utc_s and the sidecar's conversion.t_start_s give the absolute time",
644
  "rule": "row k holds the last joint command stamped at or before t_k (zero-order hold, no shift): the command issued at t_k; with joints_source next_state the measured joints of row k+1, the position the arm reached next",
645
  "stamp_meaning": "command_sent: header.stamp of the command_sent message, when the follower's motion call (set_servo_angle, or set_servo_angle_j in servo mode) returned; target_joints: header.stamp of the leader's target; next_state: the next row's slot time",
646
  "execution": "the follower validates each target and commands the arm from it; how depends on the arm's control mode, named per command by command_sent.method (per arm command_methods, per episode the sidecar's conversion.streams.command_sent.<side>.methods) and described in execution_by_method. The issue time of each command is header.stamp of the command_sent stream",
@@ -653,7 +653,7 @@
653
  ]
654
  },
655
  "execution_by_method": {
656
- "set_servo_angle_j": "servo mode (xArm mode 1, ADR-0177): the follower takes each accepted target as joints (its own solution of a TCP target, or a leader's joint target), steps a joint setpoint toward the newest one on a fixed tick (rates.control.servo_rate_hz), each joint inside the stream's speed and acceleration caps, and sends every setpoint with set_servo_angle_j; the control box executes the last setpoint it received and plans nothing. command_sent has one entry per setpoint: joints_rad is the setpoint, the pose and target_stamp are those of the target it steps toward"
657
  },
658
  "action_hold_s": 0.2,
659
  "valid_mask": "paddy.action_valid",
@@ -699,12 +699,12 @@
699
  "left": "urdf_fk",
700
  "right": "urdf_fk"
701
  },
702
- "sources": "urdf_fk: the recorded URDF's base_T_tcp at the recorded joint states (100 Hz), the kinematics the wrist cameras' paddy.camera_pose and the teleop targets use (ADR-0182); reported: the pose the arm's control box reports (about 50 Hz, its own kinematics, which can differ from the URDF's by several mm: sidecar conversion.kinematics.<side>.tcp_vs_urdf_mm); mixed: episodes differ, see each sidecar's conversion.tcp_pose.<side>",
703
- "stamps": "urdf_fk: the joint states' stamps. reported: the control box reports the pose ahead of the arm (ADR-0175); each sample is restamped by that lead, measured per episode against the URDF's TCP at the recorded joints, else the rig profile's actual_pose_lead_ms: sidecar conversion.eef_timing.<side>"
704
  },
705
  "eef_state": {
706
  "feature": "observation.eef_state",
707
- "meaning": "every arm's measured TCP pose (observation.eef_pose.<side>, sign-continuous) and gripper opening (observation.state's) in one vector, the measured counterpart of a tcp action: the state of an end-effector policy (ADR-0167)",
708
  "layout_per_arm": [
709
  "x",
710
  "y",
@@ -723,29 +723,29 @@
723
  },
724
  "kinematics": {
725
  "sidecar": "conversion.kinematics.<side> of meta/paddy/episodes/<trial_id>.json",
726
- "tcp_vs_urdf": "tcp_vs_urdf_mm and tcp_vs_urdf_deg (median, max): the URDF's base_T_tcp at the recorded joints against the recorded TCP pose (ADR-0078)",
727
- "dh": "dh: the arm control box's own DH table (metadata.json controller_settings.<side>.dh_params: 28 values, four per joint, mm and rad) as identified against the recorded TCP pose: layout (the name of each of the four values per joint, in the table's order), convention (modified = Craig, or standard), tcp_vs_dh_mm and tcp_vs_dh_deg (median, max), samples, fit, and dh_vs_urdf_mm (the table's flange against the URDF's at the same joints); with fit true, forward kinematics of the table read that way is the arm's calibrated kinematics. {note} when the trial has no table (ADR-0131)"
728
  },
729
  "cameras": {
730
  "file": "meta/cameras.json",
731
- "role": "where the camera is mounted, as the rig profile declares it (ADR-0184): overhead (fixed, looking down at the workspace), scene (fixed, another viewpoint) or wrist (moves with an arm; which arm is mount.side); null = not declared",
732
  "primary": "the file's top-level primary names the rig's one primary camera (the view to look at first), and that camera's entry has primary: true; null = not declared",
733
  "base_T_cam": "4x4 pose of the camera's color optical frame (z forward, x right, y down) in each arm base frame, metres; extrinsics.<base>.source says where it came from",
734
  "sources": {
735
  "tf_static": "the nominal pose: the rig's URDF and the camera mounts in the rig profile (CAD or tape), as published on /tf_static",
736
  "tf_snapshot": "the nominal pose from the first /tf snapshot at episode start (valid: episode_start)",
737
- "urdf_fk": "the nominal pose of a wrist camera by forward kinematics of the recorded URDF and joint states at t_s, the episode's first slot (valid: episode_start, ADR-0078)",
738
- "measured": "the rig profile's measured calibration (ADR-0132): calibration.frame_T_cam chained to the base through via (identity, tf_static, urdf, urdf_fk at t_s for a wrist camera, or tf_snapshot); the nominal entry is kept as nominal {base_T_cam, source}"
739
  },
740
  "measured_vs_nominal": "measured_vs_nominal_mm / _deg: distance of the origins and angle of the rotation between the measured and the nominal base_T_cam at the episode's first slot",
741
- "intrinsics": "K (row-major 3x3) and D (plumb_bob k1 k2 p1 p2 k3) describe the raw color image as recorded (not undistorted). intrinsics.source camera_info: the camera's own CameraInfo (a RealSense's factory values; a D435 color stream reports D all zeros). measured: the rig profile's measured intrinsics (ADR-0181: method, calibrated_at, error_px, samples, notes, from), the CameraInfo values kept as factory_intrinsics {K, D, distortion_model}. The pose of a fixed camera measured after its intrinsics were is solved with them",
742
  "calibration": "per camera: the measured calibration (frame, xyz_mm, rpy_deg URDF fixed-axis, frame_T_cam, method, calibrated_at, error_mm, error_px, samples, notes, from: rig_snapshot | rig_option), null when the rig profile has none. TF on the rig stays nominal",
743
- "image_rotation": "image_rotation_deg per camera (ADR-0194): 0, or 180 for a camera mounted upside down whose pictures the dataset turns. Turned, the color and depth frames are the recorded ones turned by 180 degrees and every value that describes them is of the turned image: K (cx' = W - 1 - cx, cy' = H - 1 - cy), D (p1 and p2 negated), base_T_cam, mount link_T_cam and tcp_T_cam, calibration.frame_T_cam, the depth K, D and color_T_depth, and paddy.camera_pose.<camera>: the camera's own optical frame (frame_id) turned about its z axis, X_T_turned = X_T_cam @ diag(-1, -1, 1, 1). calibration.xyz_mm and rpy_deg stay the rig profile's record of the camera's own optical frame. A policy trained on it needs the live frames turned the same way (paddy's policy server does it from this file)",
744
- "per_row_pose": "paddy.camera_pose.<camera> of a wrist camera; source measured_fk (forward kinematics to the calibration frame times its measured pose), urdf_fk or eef; time row (at the joint sample observation.state holds, so it matches observation.state through the URDF; the image is paddy.frame_age.<camera> older) or frame (at the shown frame's stamp), ADR-0179"
745
  },
746
  "subtasks": {
747
  "feature": "paddy.subtask_index",
748
- "meaning": "the operator-marked subtask open at the row (metadata.json subtasks, ADR-0133): its index, -1 where none is open. Spans are half-open [start, end) in header.stamp POSIX seconds; paused spans are out of the grid, so a subtask with a pause inside has no rows for the paused time",
749
  "names": "the episode sidecar's conversion.subtasks: per index the plain-language instruction, the outcome (success, failure or unknown), start_s, end_s (null: still open at the end of the recording) and the rows [start_row, end_row) it covers (null when none)",
750
  "not_a_policy_input": true
751
  },
@@ -810,12 +810,12 @@
810
  "frames": {
811
  "right": "Right_link_eef"
812
  },
813
- "frame_note": "the flange (URDF link_eef), which the six-axis sensor is mounted on (ADR-0080). Checked on the rig (validation 1.26, 2026-11-07): a press along each flange axis reads on that axis with the sign of the force applied to the tool (a push toward flange +x reads +fx); a load fit put the sensor axes within 1.4 deg of the flange's",
814
  "reading": "observation.force_torque.<side>: the control box's reading after its load and offset compensation (/paddy/<side>_arm/ft_wrench, SDK ft_ext_force): near zero, up to a small offset, with nothing touching the tool when the payload set on the control box matches the tool. paddy.force_torque_raw.<side>: the unprocessed reading (/paddy/<side>_arm/ft_wrench_raw, SDK ft_raw_force), so the compensation can be redone",
815
  "compensation": "the payload the control box compensated for is in each episode sidecar's metadata.controller_settings.<side>.tcp_load_kg and tcp_load_cog_mm, read at the trial's start",
816
- "timing": "row k holds the last reading stamped at or before t_k (zero-order hold, fresh within hold_max_age_s); the stamp is when the follower's host read the control box's real-time report packet (ADR-0081)",
817
  "missing": "rows without a fresh reading hold zeros and paddy.force_torque_valid false; they are not dropped. An arm whose rig declares the sensor but recorded no readings has the columns, all zeros and invalid",
818
- "rate_hz_nominal": "about 100 (the real-time report); 5 when the follower fell back to the rich report (ADR-0081)",
819
  "measured_median_hz": {
820
  "right": 99.5
821
  },
@@ -868,7 +868,7 @@
868
  "action": {
869
  "feature": "action",
870
  "space": "joints",
871
- "meaning": "per arm the joint command, zero-order held, then the gripper command: the quantity a policy that drives the arms through target_joints outputs (ADR-0161). The TCP target the teleop controller published is paddy.target_pose.<side>; the pose the follower sent is paddy.command_sent.<side>",
872
  "joints": "absolute positions of the URDF joints joint1..jointN, base to wrist, in observation.state's order; not a delta, not a TCP pose",
873
  "joint_unit": "rad",
874
  "joints_source": {
@@ -877,7 +877,7 @@
877
  },
878
  "joints_sources": {
879
  "command_sent": "the joint angles the follower sent (command_sent joints_rad: a set_servo_angle call's target or, in servo mode, a set_servo_angle_j setpoint; the xArm7 joint path, GELLO and policy arms), held with action_hold_s; timing.command_methods says which",
880
- "target_joints": "a leader arm's recorded joint targets (target_joints, GELLO, ADR-0097), held with action_hold_s",
881
  "next_state": "the next row's measured joints (the last row its own): the usual stand-in for a position-controlled arm's joint command when none was recorded (a mode-7 arm, older bags)"
882
  },
883
  "joints_source_per_episode": "conversion.action_source.<side> of meta/paddy/episodes/<trial_id>.json",
@@ -938,10 +938,10 @@
938
  }
939
  },
940
  "already_scaled": true,
941
- "scaling_note": "action is in robot units after the controller's motion scaling and rate limit (ADR-0046); paddy.scaling is the human-input gain in force (informational, not a policy input)"
942
  },
943
  "timing": {
944
- "row_time": "t_k = t0 + k / fps (t0 = the latest first stamp over the arms' joint states and the cameras' color frames, ADR-0173; paused spans removed); timestamp = k / fps; paddy.utc_s and the sidecar's conversion.t_start_s give the absolute time",
945
  "rule": "row k holds the last joint command stamped at or before t_k (zero-order hold, no shift): the command issued at t_k; with joints_source next_state the measured joints of row k+1, the position the arm reached next",
946
  "stamp_meaning": "command_sent: header.stamp of the command_sent message, when the follower's motion call (set_servo_angle, or set_servo_angle_j in servo mode) returned; target_joints: header.stamp of the leader's target; next_state: the next row's slot time",
947
  "execution": "the follower validates each target and commands the arm from it; how depends on the arm's control mode, named per command by command_sent.method (per arm command_methods, per episode the sidecar's conversion.streams.command_sent.<side>.methods) and described in execution_by_method. The issue time of each command is header.stamp of the command_sent stream",
@@ -954,7 +954,7 @@
954
  ]
955
  },
956
  "execution_by_method": {
957
- "set_servo_angle_j": "servo mode (xArm mode 1, ADR-0177): the follower takes each accepted target as joints (its own solution of a TCP target, or a leader's joint target), steps a joint setpoint toward the newest one on a fixed tick (rates.control.servo_rate_hz), each joint inside the stream's speed and acceleration caps, and sends every setpoint with set_servo_angle_j; the control box executes the last setpoint it received and plans nothing. command_sent has one entry per setpoint: joints_rad is the setpoint, the pose and target_stamp are those of the target it steps toward"
958
  },
959
  "action_hold_s": 0.2,
960
  "valid_mask": "paddy.action_valid",
@@ -1000,12 +1000,12 @@
1000
  "left": "urdf_fk",
1001
  "right": "urdf_fk"
1002
  },
1003
- "sources": "urdf_fk: the recorded URDF's base_T_tcp at the recorded joint states (100 Hz), the kinematics the wrist cameras' paddy.camera_pose and the teleop targets use (ADR-0182); reported: the pose the arm's control box reports (about 50 Hz, its own kinematics, which can differ from the URDF's by several mm: sidecar conversion.kinematics.<side>.tcp_vs_urdf_mm); mixed: episodes differ, see each sidecar's conversion.tcp_pose.<side>",
1004
- "stamps": "urdf_fk: the joint states' stamps. reported: the control box reports the pose ahead of the arm (ADR-0175); each sample is restamped by that lead, measured per episode against the URDF's TCP at the recorded joints, else the rig profile's actual_pose_lead_ms: sidecar conversion.eef_timing.<side>"
1005
  },
1006
  "eef_state": {
1007
  "feature": "observation.eef_state",
1008
- "meaning": "every arm's measured TCP pose (observation.eef_pose.<side>, sign-continuous) and gripper opening (observation.state's) in one vector, the measured counterpart of a tcp action: the state of an end-effector policy (ADR-0167)",
1009
  "layout_per_arm": [
1010
  "x",
1011
  "y",
@@ -1024,29 +1024,29 @@
1024
  },
1025
  "kinematics": {
1026
  "sidecar": "conversion.kinematics.<side> of meta/paddy/episodes/<trial_id>.json",
1027
- "tcp_vs_urdf": "tcp_vs_urdf_mm and tcp_vs_urdf_deg (median, max): the URDF's base_T_tcp at the recorded joints against the recorded TCP pose (ADR-0078)",
1028
- "dh": "dh: the arm control box's own DH table (metadata.json controller_settings.<side>.dh_params: 28 values, four per joint, mm and rad) as identified against the recorded TCP pose: layout (the name of each of the four values per joint, in the table's order), convention (modified = Craig, or standard), tcp_vs_dh_mm and tcp_vs_dh_deg (median, max), samples, fit, and dh_vs_urdf_mm (the table's flange against the URDF's at the same joints); with fit true, forward kinematics of the table read that way is the arm's calibrated kinematics. {note} when the trial has no table (ADR-0131)"
1029
  },
1030
  "cameras": {
1031
  "file": "meta/cameras.json",
1032
- "role": "where the camera is mounted, as the rig profile declares it (ADR-0184): overhead (fixed, looking down at the workspace), scene (fixed, another viewpoint) or wrist (moves with an arm; which arm is mount.side); null = not declared",
1033
  "primary": "the file's top-level primary names the rig's one primary camera (the view to look at first), and that camera's entry has primary: true; null = not declared",
1034
  "base_T_cam": "4x4 pose of the camera's color optical frame (z forward, x right, y down) in each arm base frame, metres; extrinsics.<base>.source says where it came from",
1035
  "sources": {
1036
  "tf_static": "the nominal pose: the rig's URDF and the camera mounts in the rig profile (CAD or tape), as published on /tf_static",
1037
  "tf_snapshot": "the nominal pose from the first /tf snapshot at episode start (valid: episode_start)",
1038
- "urdf_fk": "the nominal pose of a wrist camera by forward kinematics of the recorded URDF and joint states at t_s, the episode's first slot (valid: episode_start, ADR-0078)",
1039
- "measured": "the rig profile's measured calibration (ADR-0132): calibration.frame_T_cam chained to the base through via (identity, tf_static, urdf, urdf_fk at t_s for a wrist camera, or tf_snapshot); the nominal entry is kept as nominal {base_T_cam, source}"
1040
  },
1041
  "measured_vs_nominal": "measured_vs_nominal_mm / _deg: distance of the origins and angle of the rotation between the measured and the nominal base_T_cam at the episode's first slot",
1042
- "intrinsics": "K (row-major 3x3) and D (plumb_bob k1 k2 p1 p2 k3) describe the raw color image as recorded (not undistorted). intrinsics.source camera_info: the camera's own CameraInfo (a RealSense's factory values; a D435 color stream reports D all zeros). measured: the rig profile's measured intrinsics (ADR-0181: method, calibrated_at, error_px, samples, notes, from), the CameraInfo values kept as factory_intrinsics {K, D, distortion_model}. The pose of a fixed camera measured after its intrinsics were is solved with them",
1043
  "calibration": "per camera: the measured calibration (frame, xyz_mm, rpy_deg URDF fixed-axis, frame_T_cam, method, calibrated_at, error_mm, error_px, samples, notes, from: rig_snapshot | rig_option), null when the rig profile has none. TF on the rig stays nominal",
1044
- "image_rotation": "image_rotation_deg per camera (ADR-0194): 0, or 180 for a camera mounted upside down whose pictures the dataset turns. Turned, the color and depth frames are the recorded ones turned by 180 degrees and every value that describes them is of the turned image: K (cx' = W - 1 - cx, cy' = H - 1 - cy), D (p1 and p2 negated), base_T_cam, mount link_T_cam and tcp_T_cam, calibration.frame_T_cam, the depth K, D and color_T_depth, and paddy.camera_pose.<camera>: the camera's own optical frame (frame_id) turned about its z axis, X_T_turned = X_T_cam @ diag(-1, -1, 1, 1). calibration.xyz_mm and rpy_deg stay the rig profile's record of the camera's own optical frame. A policy trained on it needs the live frames turned the same way (paddy's policy server does it from this file)",
1045
- "per_row_pose": "paddy.camera_pose.<camera> of a wrist camera; source measured_fk (forward kinematics to the calibration frame times its measured pose), urdf_fk or eef; time row (at the joint sample observation.state holds, so it matches observation.state through the URDF; the image is paddy.frame_age.<camera> older) or frame (at the shown frame's stamp), ADR-0179"
1046
  },
1047
  "subtasks": {
1048
  "feature": "paddy.subtask_index",
1049
- "meaning": "the operator-marked subtask open at the row (metadata.json subtasks, ADR-0133): its index, -1 where none is open. Spans are half-open [start, end) in header.stamp POSIX seconds; paused spans are out of the grid, so a subtask with a pause inside has no rows for the paused time",
1050
  "names": "the episode sidecar's conversion.subtasks: per index the plain-language instruction, the outcome (success, failure or unknown), start_s, end_s (null: still open at the end of the recording) and the rows [start_row, end_row) it covers (null when none)",
1051
  "not_a_policy_input": true
1052
  },
@@ -1111,12 +1111,12 @@
1111
  "frames": {
1112
  "right": "Right_link_eef"
1113
  },
1114
- "frame_note": "the flange (URDF link_eef), which the six-axis sensor is mounted on (ADR-0080). Checked on the rig (validation 1.26, 2026-11-07): a press along each flange axis reads on that axis with the sign of the force applied to the tool (a push toward flange +x reads +fx); a load fit put the sensor axes within 1.4 deg of the flange's",
1115
  "reading": "observation.force_torque.<side>: the control box's reading after its load and offset compensation (/paddy/<side>_arm/ft_wrench, SDK ft_ext_force): near zero, up to a small offset, with nothing touching the tool when the payload set on the control box matches the tool. paddy.force_torque_raw.<side>: the unprocessed reading (/paddy/<side>_arm/ft_wrench_raw, SDK ft_raw_force), so the compensation can be redone",
1116
  "compensation": "the payload the control box compensated for is in each episode sidecar's metadata.controller_settings.<side>.tcp_load_kg and tcp_load_cog_mm, read at the trial's start",
1117
- "timing": "row k holds the last reading stamped at or before t_k (zero-order hold, fresh within hold_max_age_s); the stamp is when the follower's host read the control box's real-time report packet (ADR-0081)",
1118
  "missing": "rows without a fresh reading hold zeros and paddy.force_torque_valid false; they are not dropped. An arm whose rig declares the sensor but recorded no readings has the columns, all zeros and invalid",
1119
- "rate_hz_nominal": "about 100 (the real-time report); 5 when the follower fell back to the rich report (ADR-0081)",
1120
  "measured_median_hz": {
1121
  "right": 99.5
1122
  },
@@ -1169,7 +1169,7 @@
1169
  "action": {
1170
  "feature": "action",
1171
  "space": "joints",
1172
- "meaning": "per arm the joint command, zero-order held, then the gripper command: the quantity a policy that drives the arms through target_joints outputs (ADR-0161). The TCP target the teleop controller published is paddy.target_pose.<side>; the pose the follower sent is paddy.command_sent.<side>",
1173
  "joints": "absolute positions of the URDF joints joint1..jointN, base to wrist, in observation.state's order; not a delta, not a TCP pose",
1174
  "joint_unit": "rad",
1175
  "joints_source": {
@@ -1178,7 +1178,7 @@
1178
  },
1179
  "joints_sources": {
1180
  "command_sent": "the joint angles the follower sent (command_sent joints_rad: a set_servo_angle call's target or, in servo mode, a set_servo_angle_j setpoint; the xArm7 joint path, GELLO and policy arms), held with action_hold_s; timing.command_methods says which",
1181
- "target_joints": "a leader arm's recorded joint targets (target_joints, GELLO, ADR-0097), held with action_hold_s",
1182
  "next_state": "the next row's measured joints (the last row its own): the usual stand-in for a position-controlled arm's joint command when none was recorded (a mode-7 arm, older bags)"
1183
  },
1184
  "joints_source_per_episode": "conversion.action_source.<side> of meta/paddy/episodes/<trial_id>.json",
@@ -1239,10 +1239,10 @@
1239
  }
1240
  },
1241
  "already_scaled": true,
1242
- "scaling_note": "action is in robot units after the controller's motion scaling and rate limit (ADR-0046); paddy.scaling is the human-input gain in force (informational, not a policy input)"
1243
  },
1244
  "timing": {
1245
- "row_time": "t_k = t0 + k / fps (t0 = the latest first stamp over the arms' joint states and the cameras' color frames, ADR-0173; paused spans removed); timestamp = k / fps; paddy.utc_s and the sidecar's conversion.t_start_s give the absolute time",
1246
  "rule": "row k holds the last joint command stamped at or before t_k (zero-order hold, no shift): the command issued at t_k; with joints_source next_state the measured joints of row k+1, the position the arm reached next",
1247
  "stamp_meaning": "command_sent: header.stamp of the command_sent message, when the follower's motion call (set_servo_angle, or set_servo_angle_j in servo mode) returned; target_joints: header.stamp of the leader's target; next_state: the next row's slot time",
1248
  "execution": "the follower validates each target and commands the arm from it; how depends on the arm's control mode, named per command by command_sent.method (per arm command_methods, per episode the sidecar's conversion.streams.command_sent.<side>.methods) and described in execution_by_method. The issue time of each command is header.stamp of the command_sent stream",
@@ -1255,7 +1255,7 @@
1255
  ]
1256
  },
1257
  "execution_by_method": {
1258
- "set_servo_angle_j": "servo mode (xArm mode 1, ADR-0177): the follower takes each accepted target as joints (its own solution of a TCP target, or a leader's joint target), steps a joint setpoint toward the newest one on a fixed tick (rates.control.servo_rate_hz), each joint inside the stream's speed and acceleration caps, and sends every setpoint with set_servo_angle_j; the control box executes the last setpoint it received and plans nothing. command_sent has one entry per setpoint: joints_rad is the setpoint, the pose and target_stamp are those of the target it steps toward"
1259
  },
1260
  "action_hold_s": 0.2,
1261
  "valid_mask": "paddy.action_valid",
@@ -1301,12 +1301,12 @@
1301
  "left": "urdf_fk",
1302
  "right": "urdf_fk"
1303
  },
1304
- "sources": "urdf_fk: the recorded URDF's base_T_tcp at the recorded joint states (100 Hz), the kinematics the wrist cameras' paddy.camera_pose and the teleop targets use (ADR-0182); reported: the pose the arm's control box reports (about 50 Hz, its own kinematics, which can differ from the URDF's by several mm: sidecar conversion.kinematics.<side>.tcp_vs_urdf_mm); mixed: episodes differ, see each sidecar's conversion.tcp_pose.<side>",
1305
- "stamps": "urdf_fk: the joint states' stamps. reported: the control box reports the pose ahead of the arm (ADR-0175); each sample is restamped by that lead, measured per episode against the URDF's TCP at the recorded joints, else the rig profile's actual_pose_lead_ms: sidecar conversion.eef_timing.<side>"
1306
  },
1307
  "eef_state": {
1308
  "feature": "observation.eef_state",
1309
- "meaning": "every arm's measured TCP pose (observation.eef_pose.<side>, sign-continuous) and gripper opening (observation.state's) in one vector, the measured counterpart of a tcp action: the state of an end-effector policy (ADR-0167)",
1310
  "layout_per_arm": [
1311
  "x",
1312
  "y",
@@ -1325,29 +1325,29 @@
1325
  },
1326
  "kinematics": {
1327
  "sidecar": "conversion.kinematics.<side> of meta/paddy/episodes/<trial_id>.json",
1328
- "tcp_vs_urdf": "tcp_vs_urdf_mm and tcp_vs_urdf_deg (median, max): the URDF's base_T_tcp at the recorded joints against the recorded TCP pose (ADR-0078)",
1329
- "dh": "dh: the arm control box's own DH table (metadata.json controller_settings.<side>.dh_params: 28 values, four per joint, mm and rad) as identified against the recorded TCP pose: layout (the name of each of the four values per joint, in the table's order), convention (modified = Craig, or standard), tcp_vs_dh_mm and tcp_vs_dh_deg (median, max), samples, fit, and dh_vs_urdf_mm (the table's flange against the URDF's at the same joints); with fit true, forward kinematics of the table read that way is the arm's calibrated kinematics. {note} when the trial has no table (ADR-0131)"
1330
  },
1331
  "cameras": {
1332
  "file": "meta/cameras.json",
1333
- "role": "where the camera is mounted, as the rig profile declares it (ADR-0184): overhead (fixed, looking down at the workspace), scene (fixed, another viewpoint) or wrist (moves with an arm; which arm is mount.side); null = not declared",
1334
  "primary": "the file's top-level primary names the rig's one primary camera (the view to look at first), and that camera's entry has primary: true; null = not declared",
1335
  "base_T_cam": "4x4 pose of the camera's color optical frame (z forward, x right, y down) in each arm base frame, metres; extrinsics.<base>.source says where it came from",
1336
  "sources": {
1337
  "tf_static": "the nominal pose: the rig's URDF and the camera mounts in the rig profile (CAD or tape), as published on /tf_static",
1338
  "tf_snapshot": "the nominal pose from the first /tf snapshot at episode start (valid: episode_start)",
1339
- "urdf_fk": "the nominal pose of a wrist camera by forward kinematics of the recorded URDF and joint states at t_s, the episode's first slot (valid: episode_start, ADR-0078)",
1340
- "measured": "the rig profile's measured calibration (ADR-0132): calibration.frame_T_cam chained to the base through via (identity, tf_static, urdf, urdf_fk at t_s for a wrist camera, or tf_snapshot); the nominal entry is kept as nominal {base_T_cam, source}"
1341
  },
1342
  "measured_vs_nominal": "measured_vs_nominal_mm / _deg: distance of the origins and angle of the rotation between the measured and the nominal base_T_cam at the episode's first slot",
1343
- "intrinsics": "K (row-major 3x3) and D (plumb_bob k1 k2 p1 p2 k3) describe the raw color image as recorded (not undistorted). intrinsics.source camera_info: the camera's own CameraInfo (a RealSense's factory values; a D435 color stream reports D all zeros). measured: the rig profile's measured intrinsics (ADR-0181: method, calibrated_at, error_px, samples, notes, from), the CameraInfo values kept as factory_intrinsics {K, D, distortion_model}. The pose of a fixed camera measured after its intrinsics were is solved with them",
1344
  "calibration": "per camera: the measured calibration (frame, xyz_mm, rpy_deg URDF fixed-axis, frame_T_cam, method, calibrated_at, error_mm, error_px, samples, notes, from: rig_snapshot | rig_option), null when the rig profile has none. TF on the rig stays nominal",
1345
- "image_rotation": "image_rotation_deg per camera (ADR-0194): 0, or 180 for a camera mounted upside down whose pictures the dataset turns. Turned, the color and depth frames are the recorded ones turned by 180 degrees and every value that describes them is of the turned image: K (cx' = W - 1 - cx, cy' = H - 1 - cy), D (p1 and p2 negated), base_T_cam, mount link_T_cam and tcp_T_cam, calibration.frame_T_cam, the depth K, D and color_T_depth, and paddy.camera_pose.<camera>: the camera's own optical frame (frame_id) turned about its z axis, X_T_turned = X_T_cam @ diag(-1, -1, 1, 1). calibration.xyz_mm and rpy_deg stay the rig profile's record of the camera's own optical frame. A policy trained on it needs the live frames turned the same way (paddy's policy server does it from this file)",
1346
- "per_row_pose": "paddy.camera_pose.<camera> of a wrist camera; source measured_fk (forward kinematics to the calibration frame times its measured pose), urdf_fk or eef; time row (at the joint sample observation.state holds, so it matches observation.state through the URDF; the image is paddy.frame_age.<camera> older) or frame (at the shown frame's stamp), ADR-0179"
1347
  },
1348
  "subtasks": {
1349
  "feature": "paddy.subtask_index",
1350
- "meaning": "the operator-marked subtask open at the row (metadata.json subtasks, ADR-0133): its index, -1 where none is open. Spans are half-open [start, end) in header.stamp POSIX seconds; paused spans are out of the grid, so a subtask with a pause inside has no rows for the paused time",
1351
  "names": "the episode sidecar's conversion.subtasks: per index the plain-language instruction, the outcome (success, failure or unknown), start_s, end_s (null: still open at the end of the recording) and the rows [start_row, end_row) it covers (null when none)",
1352
  "not_a_policy_input": true
1353
  },
@@ -1412,12 +1412,12 @@
1412
  "frames": {
1413
  "right": "Right_link_eef"
1414
  },
1415
- "frame_note": "the flange (URDF link_eef), which the six-axis sensor is mounted on (ADR-0080). Checked on the rig (validation 1.26, 2026-11-07): a press along each flange axis reads on that axis with the sign of the force applied to the tool (a push toward flange +x reads +fx); a load fit put the sensor axes within 1.4 deg of the flange's",
1416
  "reading": "observation.force_torque.<side>: the control box's reading after its load and offset compensation (/paddy/<side>_arm/ft_wrench, SDK ft_ext_force): near zero, up to a small offset, with nothing touching the tool when the payload set on the control box matches the tool. paddy.force_torque_raw.<side>: the unprocessed reading (/paddy/<side>_arm/ft_wrench_raw, SDK ft_raw_force), so the compensation can be redone",
1417
  "compensation": "the payload the control box compensated for is in each episode sidecar's metadata.controller_settings.<side>.tcp_load_kg and tcp_load_cog_mm, read at the trial's start",
1418
- "timing": "row k holds the last reading stamped at or before t_k (zero-order hold, fresh within hold_max_age_s); the stamp is when the follower's host read the control box's real-time report packet (ADR-0081)",
1419
  "missing": "rows without a fresh reading hold zeros and paddy.force_torque_valid false; they are not dropped. An arm whose rig declares the sensor but recorded no readings has the columns, all zeros and invalid",
1420
- "rate_hz_nominal": "about 100 (the real-time report); 5 when the follower fell back to the rich report (ADR-0081)",
1421
  "measured_median_hz": {
1422
  "right": 99.5
1423
  },
@@ -1470,7 +1470,7 @@
1470
  "action": {
1471
  "feature": "action",
1472
  "space": "joints",
1473
- "meaning": "per arm the joint command, zero-order held, then the gripper command: the quantity a policy that drives the arms through target_joints outputs (ADR-0161). The TCP target the teleop controller published is paddy.target_pose.<side>; the pose the follower sent is paddy.command_sent.<side>",
1474
  "joints": "absolute positions of the URDF joints joint1..jointN, base to wrist, in observation.state's order; not a delta, not a TCP pose",
1475
  "joint_unit": "rad",
1476
  "joints_source": {
@@ -1479,7 +1479,7 @@
1479
  },
1480
  "joints_sources": {
1481
  "command_sent": "the joint angles the follower sent (command_sent joints_rad: a set_servo_angle call's target or, in servo mode, a set_servo_angle_j setpoint; the xArm7 joint path, GELLO and policy arms), held with action_hold_s; timing.command_methods says which",
1482
- "target_joints": "a leader arm's recorded joint targets (target_joints, GELLO, ADR-0097), held with action_hold_s",
1483
  "next_state": "the next row's measured joints (the last row its own): the usual stand-in for a position-controlled arm's joint command when none was recorded (a mode-7 arm, older bags)"
1484
  },
1485
  "joints_source_per_episode": "conversion.action_source.<side> of meta/paddy/episodes/<trial_id>.json",
@@ -1540,10 +1540,10 @@
1540
  }
1541
  },
1542
  "already_scaled": true,
1543
- "scaling_note": "action is in robot units after the controller's motion scaling and rate limit (ADR-0046); paddy.scaling is the human-input gain in force (informational, not a policy input)"
1544
  },
1545
  "timing": {
1546
- "row_time": "t_k = t0 + k / fps (t0 = the latest first stamp over the arms' joint states and the cameras' color frames, ADR-0173; paused spans removed); timestamp = k / fps; paddy.utc_s and the sidecar's conversion.t_start_s give the absolute time",
1547
  "rule": "row k holds the last joint command stamped at or before t_k (zero-order hold, no shift): the command issued at t_k; with joints_source next_state the measured joints of row k+1, the position the arm reached next",
1548
  "stamp_meaning": "command_sent: header.stamp of the command_sent message, when the follower's motion call (set_servo_angle, or set_servo_angle_j in servo mode) returned; target_joints: header.stamp of the leader's target; next_state: the next row's slot time",
1549
  "execution": "the follower validates each target and commands the arm from it; how depends on the arm's control mode, named per command by command_sent.method (per arm command_methods, per episode the sidecar's conversion.streams.command_sent.<side>.methods) and described in execution_by_method. The issue time of each command is header.stamp of the command_sent stream",
@@ -1556,7 +1556,7 @@
1556
  ]
1557
  },
1558
  "execution_by_method": {
1559
- "set_servo_angle_j": "servo mode (xArm mode 1, ADR-0177): the follower takes each accepted target as joints (its own solution of a TCP target, or a leader's joint target), steps a joint setpoint toward the newest one on a fixed tick (rates.control.servo_rate_hz), each joint inside the stream's speed and acceleration caps, and sends every setpoint with set_servo_angle_j; the control box executes the last setpoint it received and plans nothing. command_sent has one entry per setpoint: joints_rad is the setpoint, the pose and target_stamp are those of the target it steps toward"
1560
  },
1561
  "action_hold_s": 0.2,
1562
  "valid_mask": "paddy.action_valid",
@@ -1602,12 +1602,12 @@
1602
  "left": "urdf_fk",
1603
  "right": "urdf_fk"
1604
  },
1605
- "sources": "urdf_fk: the recorded URDF's base_T_tcp at the recorded joint states (100 Hz), the kinematics the wrist cameras' paddy.camera_pose and the teleop targets use (ADR-0182); reported: the pose the arm's control box reports (about 50 Hz, its own kinematics, which can differ from the URDF's by several mm: sidecar conversion.kinematics.<side>.tcp_vs_urdf_mm); mixed: episodes differ, see each sidecar's conversion.tcp_pose.<side>",
1606
- "stamps": "urdf_fk: the joint states' stamps. reported: the control box reports the pose ahead of the arm (ADR-0175); each sample is restamped by that lead, measured per episode against the URDF's TCP at the recorded joints, else the rig profile's actual_pose_lead_ms: sidecar conversion.eef_timing.<side>"
1607
  },
1608
  "eef_state": {
1609
  "feature": "observation.eef_state",
1610
- "meaning": "every arm's measured TCP pose (observation.eef_pose.<side>, sign-continuous) and gripper opening (observation.state's) in one vector, the measured counterpart of a tcp action: the state of an end-effector policy (ADR-0167)",
1611
  "layout_per_arm": [
1612
  "x",
1613
  "y",
@@ -1626,29 +1626,29 @@
1626
  },
1627
  "kinematics": {
1628
  "sidecar": "conversion.kinematics.<side> of meta/paddy/episodes/<trial_id>.json",
1629
- "tcp_vs_urdf": "tcp_vs_urdf_mm and tcp_vs_urdf_deg (median, max): the URDF's base_T_tcp at the recorded joints against the recorded TCP pose (ADR-0078)",
1630
- "dh": "dh: the arm control box's own DH table (metadata.json controller_settings.<side>.dh_params: 28 values, four per joint, mm and rad) as identified against the recorded TCP pose: layout (the name of each of the four values per joint, in the table's order), convention (modified = Craig, or standard), tcp_vs_dh_mm and tcp_vs_dh_deg (median, max), samples, fit, and dh_vs_urdf_mm (the table's flange against the URDF's at the same joints); with fit true, forward kinematics of the table read that way is the arm's calibrated kinematics. {note} when the trial has no table (ADR-0131)"
1631
  },
1632
  "cameras": {
1633
  "file": "meta/cameras.json",
1634
- "role": "where the camera is mounted, as the rig profile declares it (ADR-0184): overhead (fixed, looking down at the workspace), scene (fixed, another viewpoint) or wrist (moves with an arm; which arm is mount.side); null = not declared",
1635
  "primary": "the file's top-level primary names the rig's one primary camera (the view to look at first), and that camera's entry has primary: true; null = not declared",
1636
  "base_T_cam": "4x4 pose of the camera's color optical frame (z forward, x right, y down) in each arm base frame, metres; extrinsics.<base>.source says where it came from",
1637
  "sources": {
1638
  "tf_static": "the nominal pose: the rig's URDF and the camera mounts in the rig profile (CAD or tape), as published on /tf_static",
1639
  "tf_snapshot": "the nominal pose from the first /tf snapshot at episode start (valid: episode_start)",
1640
- "urdf_fk": "the nominal pose of a wrist camera by forward kinematics of the recorded URDF and joint states at t_s, the episode's first slot (valid: episode_start, ADR-0078)",
1641
- "measured": "the rig profile's measured calibration (ADR-0132): calibration.frame_T_cam chained to the base through via (identity, tf_static, urdf, urdf_fk at t_s for a wrist camera, or tf_snapshot); the nominal entry is kept as nominal {base_T_cam, source}"
1642
  },
1643
  "measured_vs_nominal": "measured_vs_nominal_mm / _deg: distance of the origins and angle of the rotation between the measured and the nominal base_T_cam at the episode's first slot",
1644
- "intrinsics": "K (row-major 3x3) and D (plumb_bob k1 k2 p1 p2 k3) describe the raw color image as recorded (not undistorted). intrinsics.source camera_info: the camera's own CameraInfo (a RealSense's factory values; a D435 color stream reports D all zeros). measured: the rig profile's measured intrinsics (ADR-0181: method, calibrated_at, error_px, samples, notes, from), the CameraInfo values kept as factory_intrinsics {K, D, distortion_model}. The pose of a fixed camera measured after its intrinsics were is solved with them",
1645
  "calibration": "per camera: the measured calibration (frame, xyz_mm, rpy_deg URDF fixed-axis, frame_T_cam, method, calibrated_at, error_mm, error_px, samples, notes, from: rig_snapshot | rig_option), null when the rig profile has none. TF on the rig stays nominal",
1646
- "image_rotation": "image_rotation_deg per camera (ADR-0194): 0, or 180 for a camera mounted upside down whose pictures the dataset turns. Turned, the color and depth frames are the recorded ones turned by 180 degrees and every value that describes them is of the turned image: K (cx' = W - 1 - cx, cy' = H - 1 - cy), D (p1 and p2 negated), base_T_cam, mount link_T_cam and tcp_T_cam, calibration.frame_T_cam, the depth K, D and color_T_depth, and paddy.camera_pose.<camera>: the camera's own optical frame (frame_id) turned about its z axis, X_T_turned = X_T_cam @ diag(-1, -1, 1, 1). calibration.xyz_mm and rpy_deg stay the rig profile's record of the camera's own optical frame. A policy trained on it needs the live frames turned the same way (paddy's policy server does it from this file)",
1647
- "per_row_pose": "paddy.camera_pose.<camera> of a wrist camera; source measured_fk (forward kinematics to the calibration frame times its measured pose), urdf_fk or eef; time row (at the joint sample observation.state holds, so it matches observation.state through the URDF; the image is paddy.frame_age.<camera> older) or frame (at the shown frame's stamp), ADR-0179"
1648
  },
1649
  "subtasks": {
1650
  "feature": "paddy.subtask_index",
1651
- "meaning": "the operator-marked subtask open at the row (metadata.json subtasks, ADR-0133): its index, -1 where none is open. Spans are half-open [start, end) in header.stamp POSIX seconds; paused spans are out of the grid, so a subtask with a pause inside has no rows for the paused time",
1652
  "names": "the episode sidecar's conversion.subtasks: per index the plain-language instruction, the outcome (success, failure or unknown), start_s, end_s (null: still open at the end of the recording) and the rows [start_row, end_row) it covers (null when none)",
1653
  "not_a_policy_input": true
1654
  },
@@ -1713,12 +1713,12 @@
1713
  "frames": {
1714
  "right": "Right_link_eef"
1715
  },
1716
- "frame_note": "the flange (URDF link_eef), which the six-axis sensor is mounted on (ADR-0080). Checked on the rig (validation 1.26, 2026-11-07): a press along each flange axis reads on that axis with the sign of the force applied to the tool (a push toward flange +x reads +fx); a load fit put the sensor axes within 1.4 deg of the flange's",
1717
  "reading": "observation.force_torque.<side>: the control box's reading after its load and offset compensation (/paddy/<side>_arm/ft_wrench, SDK ft_ext_force): near zero, up to a small offset, with nothing touching the tool when the payload set on the control box matches the tool. paddy.force_torque_raw.<side>: the unprocessed reading (/paddy/<side>_arm/ft_wrench_raw, SDK ft_raw_force), so the compensation can be redone",
1718
  "compensation": "the payload the control box compensated for is in each episode sidecar's metadata.controller_settings.<side>.tcp_load_kg and tcp_load_cog_mm, read at the trial's start",
1719
- "timing": "row k holds the last reading stamped at or before t_k (zero-order hold, fresh within hold_max_age_s); the stamp is when the follower's host read the control box's real-time report packet (ADR-0081)",
1720
  "missing": "rows without a fresh reading hold zeros and paddy.force_torque_valid false; they are not dropped. An arm whose rig declares the sensor but recorded no readings has the columns, all zeros and invalid",
1721
- "rate_hz_nominal": "about 100 (the real-time report); 5 when the follower fell back to the rich report (ADR-0081)",
1722
  "measured_median_hz": {
1723
  "right": 99.5
1724
  },
 
250
  "action": {
251
  "feature": "action",
252
  "space": "joints",
253
+ "meaning": "per arm the joint command, zero-order held, then the gripper command: the quantity a policy that drives the arms through target_joints outputs. The TCP target the teleop controller published is paddy.target_pose.<side>; the pose the follower sent is paddy.command_sent.<side>",
254
  "joints": "absolute positions of the URDF joints joint1..jointN, base to wrist, in observation.state's order; not a delta, not a TCP pose",
255
  "joint_unit": "rad",
256
  "joints_source": {
 
259
  },
260
  "joints_sources": {
261
  "command_sent": "the joint angles the follower sent (command_sent joints_rad: a set_servo_angle call's target or, in servo mode, a set_servo_angle_j setpoint; the xArm7 joint path, GELLO and policy arms), held with action_hold_s; timing.command_methods says which",
262
+ "target_joints": "a leader arm's recorded joint targets (target_joints, GELLO), held with action_hold_s",
263
  "next_state": "the next row's measured joints (the last row its own): the usual stand-in for a position-controlled arm's joint command when none was recorded (a mode-7 arm, older bags)"
264
  },
265
  "joints_source_per_episode": "conversion.action_source.<side> of meta/paddy/episodes/<trial_id>.json",
 
320
  }
321
  },
322
  "already_scaled": true,
323
+ "scaling_note": "action is in robot units after the controller's motion scaling and rate limit; paddy.scaling is the human-input gain in force (informational, not a policy input)"
324
  },
325
  "timing": {
326
+ "row_time": "t_k = t0 + k / fps (t0 = the latest first stamp over the arms' joint states and the cameras' color frames; paused spans removed); timestamp = k / fps; paddy.utc_s and the sidecar's conversion.t_start_s give the absolute time",
327
  "rule": "row k holds the last joint command stamped at or before t_k (zero-order hold, no shift): the command issued at t_k; with joints_source next_state the measured joints of row k+1, the position the arm reached next",
328
  "stamp_meaning": "command_sent: header.stamp of the command_sent message, when the follower's motion call (set_servo_angle, or set_servo_angle_j in servo mode) returned; target_joints: header.stamp of the leader's target; next_state: the next row's slot time",
329
  "execution": "the follower validates each target and commands the arm from it; how depends on the arm's control mode, named per command by command_sent.method (per arm command_methods, per episode the sidecar's conversion.streams.command_sent.<side>.methods) and described in execution_by_method. The issue time of each command is header.stamp of the command_sent stream",
 
336
  ]
337
  },
338
  "execution_by_method": {
339
+ "set_servo_angle_j": "servo mode (xArm mode 1): the follower takes each accepted target as joints (its own solution of a TCP target, or a leader's joint target), steps a joint setpoint toward the newest one on a fixed tick (rates.control.servo_rate_hz), each joint inside the stream's speed and acceleration caps, and sends every setpoint with set_servo_angle_j; the control box executes the last setpoint it received and plans nothing. command_sent has one entry per setpoint: joints_rad is the setpoint, the pose and target_stamp are those of the target it steps toward"
340
  },
341
  "action_hold_s": 0.2,
342
  "valid_mask": "paddy.action_valid",
 
382
  "left": "urdf_fk",
383
  "right": "urdf_fk"
384
  },
385
+ "sources": "urdf_fk: the recorded URDF's base_T_tcp at the recorded joint states (100 Hz), the kinematics the wrist cameras' paddy.camera_pose and the teleop targets use; reported: the pose the arm's control box reports (about 50 Hz, its own kinematics, which can differ from the URDF's by several mm: sidecar conversion.kinematics.<side>.tcp_vs_urdf_mm); mixed: episodes differ, see each sidecar's conversion.tcp_pose.<side>",
386
+ "stamps": "urdf_fk: the joint states' stamps. reported: the control box reports the pose ahead of the arm; each sample is restamped by that lead, measured per episode against the URDF's TCP at the recorded joints, else the rig profile's actual_pose_lead_ms: sidecar conversion.eef_timing.<side>"
387
  },
388
  "eef_state": {
389
  "feature": "observation.eef_state",
390
+ "meaning": "every arm's measured TCP pose (observation.eef_pose.<side>, sign-continuous) and gripper opening (observation.state's) in one vector, the measured counterpart of a tcp action: the state of an end-effector policy",
391
  "layout_per_arm": [
392
  "x",
393
  "y",
 
406
  },
407
  "kinematics": {
408
  "sidecar": "conversion.kinematics.<side> of meta/paddy/episodes/<trial_id>.json",
409
+ "tcp_vs_urdf": "tcp_vs_urdf_mm and tcp_vs_urdf_deg (median, max): the URDF's base_T_tcp at the recorded joints against the recorded TCP pose",
410
+ "dh": "dh: the arm control box's own DH table (metadata.json controller_settings.<side>.dh_params: 28 values, four per joint, mm and rad) as identified against the recorded TCP pose: layout (the name of each of the four values per joint, in the table's order), convention (modified = Craig, or standard), tcp_vs_dh_mm and tcp_vs_dh_deg (median, max), samples, fit, and dh_vs_urdf_mm (the table's flange against the URDF's at the same joints); with fit true, forward kinematics of the table read that way is the arm's calibrated kinematics. {note} when the trial has no table"
411
  },
412
  "cameras": {
413
  "file": "meta/cameras.json",
414
+ "role": "where the camera is mounted, as the rig profile declares it: overhead (fixed, looking down at the workspace), scene (fixed, another viewpoint) or wrist (moves with an arm; which arm is mount.side); null = not declared",
415
  "primary": "the file's top-level primary names the rig's one primary camera (the view to look at first), and that camera's entry has primary: true; null = not declared",
416
  "base_T_cam": "4x4 pose of the camera's color optical frame (z forward, x right, y down) in each arm base frame, metres; extrinsics.<base>.source says where it came from",
417
  "sources": {
418
  "tf_static": "the nominal pose: the rig's URDF and the camera mounts in the rig profile (CAD or tape), as published on /tf_static",
419
  "tf_snapshot": "the nominal pose from the first /tf snapshot at episode start (valid: episode_start)",
420
+ "urdf_fk": "the nominal pose of a wrist camera by forward kinematics of the recorded URDF and joint states at t_s, the episode's first slot (valid: episode_start)",
421
+ "measured": "the rig profile's measured calibration: calibration.frame_T_cam chained to the base through via (identity, tf_static, urdf, urdf_fk at t_s for a wrist camera, or tf_snapshot); the nominal entry is kept as nominal {base_T_cam, source}"
422
  },
423
  "measured_vs_nominal": "measured_vs_nominal_mm / _deg: distance of the origins and angle of the rotation between the measured and the nominal base_T_cam at the episode's first slot",
424
+ "intrinsics": "K (row-major 3x3) and D (plumb_bob k1 k2 p1 p2 k3) describe the raw color image as recorded (not undistorted). intrinsics.source camera_info: the camera's own CameraInfo (a RealSense's factory values; a D435 color stream reports D all zeros). measured: the rig profile's measured intrinsics (method, calibrated_at, error_px, samples, notes, from), the CameraInfo values kept as factory_intrinsics {K, D, distortion_model}. The pose of a fixed camera measured after its intrinsics were is solved with them",
425
  "calibration": "per camera: the measured calibration (frame, xyz_mm, rpy_deg URDF fixed-axis, frame_T_cam, method, calibrated_at, error_mm, error_px, samples, notes, from: rig_snapshot | rig_option), null when the rig profile has none. TF on the rig stays nominal",
426
+ "image_rotation": "image_rotation_deg per camera: 0, or 180 for a camera mounted upside down whose pictures the dataset turns. Turned, the color and depth frames are the recorded ones turned by 180 degrees and every value that describes them is of the turned image: K (cx' = W - 1 - cx, cy' = H - 1 - cy), D (p1 and p2 negated), base_T_cam, mount link_T_cam and tcp_T_cam, calibration.frame_T_cam, the depth K, D and color_T_depth, and paddy.camera_pose.<camera>: the camera's own optical frame (frame_id) turned about its z axis, X_T_turned = X_T_cam @ diag(-1, -1, 1, 1). calibration.xyz_mm and rpy_deg stay the rig profile's record of the camera's own optical frame. A policy trained on it needs the live frames turned the same way (paddy's policy server does it from this file)",
427
+ "per_row_pose": "paddy.camera_pose.<camera> of a wrist camera; source measured_fk (forward kinematics to the calibration frame times its measured pose), urdf_fk or eef; time row (at the joint sample observation.state holds, so it matches observation.state through the URDF; the image is paddy.frame_age.<camera> older) or frame (at the shown frame's stamp)"
428
  },
429
  "subtasks": {
430
  "feature": "paddy.subtask_index",
431
+ "meaning": "the operator-marked subtask open at the row (metadata.json subtasks): its index, -1 where none is open. Spans are half-open [start, end) in header.stamp POSIX seconds; paused spans are out of the grid, so a subtask with a pause inside has no rows for the paused time",
432
  "names": "the episode sidecar's conversion.subtasks: per index the plain-language instruction, the outcome (success, failure or unknown), start_s, end_s (null: still open at the end of the recording) and the rows [start_row, end_row) it covers (null when none)",
433
  "not_a_policy_input": true
434
  },
 
493
  "frames": {
494
  "right": "Right_link_eef"
495
  },
496
+ "frame_note": "the flange (URDF link_eef), which the six-axis sensor is mounted on. Checked on the rig (validation 1.26, 2026-11-07): a press along each flange axis reads on that axis with the sign of the force applied to the tool (a push toward flange +x reads +fx); a load fit put the sensor axes within 1.4 deg of the flange's",
497
  "reading": "observation.force_torque.<side>: the control box's reading after its load and offset compensation (/paddy/<side>_arm/ft_wrench, SDK ft_ext_force): near zero, up to a small offset, with nothing touching the tool when the payload set on the control box matches the tool. paddy.force_torque_raw.<side>: the unprocessed reading (/paddy/<side>_arm/ft_wrench_raw, SDK ft_raw_force), so the compensation can be redone",
498
  "compensation": "the payload the control box compensated for is in each episode sidecar's metadata.controller_settings.<side>.tcp_load_kg and tcp_load_cog_mm, read at the trial's start",
499
+ "timing": "row k holds the last reading stamped at or before t_k (zero-order hold, fresh within hold_max_age_s); the stamp is when the follower's host read the control box's real-time report packet",
500
  "missing": "rows without a fresh reading hold zeros and paddy.force_torque_valid false; they are not dropped. An arm whose rig declares the sensor but recorded no readings has the columns, all zeros and invalid",
501
+ "rate_hz_nominal": "about 100 (the real-time report); 5 when the follower fell back to the rich report",
502
  "measured_median_hz": {
503
  "right": 99.5
504
  },
 
567
  "action": {
568
  "feature": "action",
569
  "space": "joints",
570
+ "meaning": "per arm the joint command, zero-order held, then the gripper command: the quantity a policy that drives the arms through target_joints outputs. The TCP target the teleop controller published is paddy.target_pose.<side>; the pose the follower sent is paddy.command_sent.<side>",
571
  "joints": "absolute positions of the URDF joints joint1..jointN, base to wrist, in observation.state's order; not a delta, not a TCP pose",
572
  "joint_unit": "rad",
573
  "joints_source": {
 
576
  },
577
  "joints_sources": {
578
  "command_sent": "the joint angles the follower sent (command_sent joints_rad: a set_servo_angle call's target or, in servo mode, a set_servo_angle_j setpoint; the xArm7 joint path, GELLO and policy arms), held with action_hold_s; timing.command_methods says which",
579
+ "target_joints": "a leader arm's recorded joint targets (target_joints, GELLO), held with action_hold_s",
580
  "next_state": "the next row's measured joints (the last row its own): the usual stand-in for a position-controlled arm's joint command when none was recorded (a mode-7 arm, older bags)"
581
  },
582
  "joints_source_per_episode": "conversion.action_source.<side> of meta/paddy/episodes/<trial_id>.json",
 
637
  }
638
  },
639
  "already_scaled": true,
640
+ "scaling_note": "action is in robot units after the controller's motion scaling and rate limit; paddy.scaling is the human-input gain in force (informational, not a policy input)"
641
  },
642
  "timing": {
643
+ "row_time": "t_k = t0 + k / fps (t0 = the latest first stamp over the arms' joint states and the cameras' color frames; paused spans removed); timestamp = k / fps; paddy.utc_s and the sidecar's conversion.t_start_s give the absolute time",
644
  "rule": "row k holds the last joint command stamped at or before t_k (zero-order hold, no shift): the command issued at t_k; with joints_source next_state the measured joints of row k+1, the position the arm reached next",
645
  "stamp_meaning": "command_sent: header.stamp of the command_sent message, when the follower's motion call (set_servo_angle, or set_servo_angle_j in servo mode) returned; target_joints: header.stamp of the leader's target; next_state: the next row's slot time",
646
  "execution": "the follower validates each target and commands the arm from it; how depends on the arm's control mode, named per command by command_sent.method (per arm command_methods, per episode the sidecar's conversion.streams.command_sent.<side>.methods) and described in execution_by_method. The issue time of each command is header.stamp of the command_sent stream",
 
653
  ]
654
  },
655
  "execution_by_method": {
656
+ "set_servo_angle_j": "servo mode (xArm mode 1): the follower takes each accepted target as joints (its own solution of a TCP target, or a leader's joint target), steps a joint setpoint toward the newest one on a fixed tick (rates.control.servo_rate_hz), each joint inside the stream's speed and acceleration caps, and sends every setpoint with set_servo_angle_j; the control box executes the last setpoint it received and plans nothing. command_sent has one entry per setpoint: joints_rad is the setpoint, the pose and target_stamp are those of the target it steps toward"
657
  },
658
  "action_hold_s": 0.2,
659
  "valid_mask": "paddy.action_valid",
 
699
  "left": "urdf_fk",
700
  "right": "urdf_fk"
701
  },
702
+ "sources": "urdf_fk: the recorded URDF's base_T_tcp at the recorded joint states (100 Hz), the kinematics the wrist cameras' paddy.camera_pose and the teleop targets use; reported: the pose the arm's control box reports (about 50 Hz, its own kinematics, which can differ from the URDF's by several mm: sidecar conversion.kinematics.<side>.tcp_vs_urdf_mm); mixed: episodes differ, see each sidecar's conversion.tcp_pose.<side>",
703
+ "stamps": "urdf_fk: the joint states' stamps. reported: the control box reports the pose ahead of the arm; each sample is restamped by that lead, measured per episode against the URDF's TCP at the recorded joints, else the rig profile's actual_pose_lead_ms: sidecar conversion.eef_timing.<side>"
704
  },
705
  "eef_state": {
706
  "feature": "observation.eef_state",
707
+ "meaning": "every arm's measured TCP pose (observation.eef_pose.<side>, sign-continuous) and gripper opening (observation.state's) in one vector, the measured counterpart of a tcp action: the state of an end-effector policy",
708
  "layout_per_arm": [
709
  "x",
710
  "y",
 
723
  },
724
  "kinematics": {
725
  "sidecar": "conversion.kinematics.<side> of meta/paddy/episodes/<trial_id>.json",
726
+ "tcp_vs_urdf": "tcp_vs_urdf_mm and tcp_vs_urdf_deg (median, max): the URDF's base_T_tcp at the recorded joints against the recorded TCP pose",
727
+ "dh": "dh: the arm control box's own DH table (metadata.json controller_settings.<side>.dh_params: 28 values, four per joint, mm and rad) as identified against the recorded TCP pose: layout (the name of each of the four values per joint, in the table's order), convention (modified = Craig, or standard), tcp_vs_dh_mm and tcp_vs_dh_deg (median, max), samples, fit, and dh_vs_urdf_mm (the table's flange against the URDF's at the same joints); with fit true, forward kinematics of the table read that way is the arm's calibrated kinematics. {note} when the trial has no table"
728
  },
729
  "cameras": {
730
  "file": "meta/cameras.json",
731
+ "role": "where the camera is mounted, as the rig profile declares it: overhead (fixed, looking down at the workspace), scene (fixed, another viewpoint) or wrist (moves with an arm; which arm is mount.side); null = not declared",
732
  "primary": "the file's top-level primary names the rig's one primary camera (the view to look at first), and that camera's entry has primary: true; null = not declared",
733
  "base_T_cam": "4x4 pose of the camera's color optical frame (z forward, x right, y down) in each arm base frame, metres; extrinsics.<base>.source says where it came from",
734
  "sources": {
735
  "tf_static": "the nominal pose: the rig's URDF and the camera mounts in the rig profile (CAD or tape), as published on /tf_static",
736
  "tf_snapshot": "the nominal pose from the first /tf snapshot at episode start (valid: episode_start)",
737
+ "urdf_fk": "the nominal pose of a wrist camera by forward kinematics of the recorded URDF and joint states at t_s, the episode's first slot (valid: episode_start)",
738
+ "measured": "the rig profile's measured calibration: calibration.frame_T_cam chained to the base through via (identity, tf_static, urdf, urdf_fk at t_s for a wrist camera, or tf_snapshot); the nominal entry is kept as nominal {base_T_cam, source}"
739
  },
740
  "measured_vs_nominal": "measured_vs_nominal_mm / _deg: distance of the origins and angle of the rotation between the measured and the nominal base_T_cam at the episode's first slot",
741
+ "intrinsics": "K (row-major 3x3) and D (plumb_bob k1 k2 p1 p2 k3) describe the raw color image as recorded (not undistorted). intrinsics.source camera_info: the camera's own CameraInfo (a RealSense's factory values; a D435 color stream reports D all zeros). measured: the rig profile's measured intrinsics (method, calibrated_at, error_px, samples, notes, from), the CameraInfo values kept as factory_intrinsics {K, D, distortion_model}. The pose of a fixed camera measured after its intrinsics were is solved with them",
742
  "calibration": "per camera: the measured calibration (frame, xyz_mm, rpy_deg URDF fixed-axis, frame_T_cam, method, calibrated_at, error_mm, error_px, samples, notes, from: rig_snapshot | rig_option), null when the rig profile has none. TF on the rig stays nominal",
743
+ "image_rotation": "image_rotation_deg per camera: 0, or 180 for a camera mounted upside down whose pictures the dataset turns. Turned, the color and depth frames are the recorded ones turned by 180 degrees and every value that describes them is of the turned image: K (cx' = W - 1 - cx, cy' = H - 1 - cy), D (p1 and p2 negated), base_T_cam, mount link_T_cam and tcp_T_cam, calibration.frame_T_cam, the depth K, D and color_T_depth, and paddy.camera_pose.<camera>: the camera's own optical frame (frame_id) turned about its z axis, X_T_turned = X_T_cam @ diag(-1, -1, 1, 1). calibration.xyz_mm and rpy_deg stay the rig profile's record of the camera's own optical frame. A policy trained on it needs the live frames turned the same way (paddy's policy server does it from this file)",
744
+ "per_row_pose": "paddy.camera_pose.<camera> of a wrist camera; source measured_fk (forward kinematics to the calibration frame times its measured pose), urdf_fk or eef; time row (at the joint sample observation.state holds, so it matches observation.state through the URDF; the image is paddy.frame_age.<camera> older) or frame (at the shown frame's stamp)"
745
  },
746
  "subtasks": {
747
  "feature": "paddy.subtask_index",
748
+ "meaning": "the operator-marked subtask open at the row (metadata.json subtasks): its index, -1 where none is open. Spans are half-open [start, end) in header.stamp POSIX seconds; paused spans are out of the grid, so a subtask with a pause inside has no rows for the paused time",
749
  "names": "the episode sidecar's conversion.subtasks: per index the plain-language instruction, the outcome (success, failure or unknown), start_s, end_s (null: still open at the end of the recording) and the rows [start_row, end_row) it covers (null when none)",
750
  "not_a_policy_input": true
751
  },
 
810
  "frames": {
811
  "right": "Right_link_eef"
812
  },
813
+ "frame_note": "the flange (URDF link_eef), which the six-axis sensor is mounted on. Checked on the rig (validation 1.26, 2026-11-07): a press along each flange axis reads on that axis with the sign of the force applied to the tool (a push toward flange +x reads +fx); a load fit put the sensor axes within 1.4 deg of the flange's",
814
  "reading": "observation.force_torque.<side>: the control box's reading after its load and offset compensation (/paddy/<side>_arm/ft_wrench, SDK ft_ext_force): near zero, up to a small offset, with nothing touching the tool when the payload set on the control box matches the tool. paddy.force_torque_raw.<side>: the unprocessed reading (/paddy/<side>_arm/ft_wrench_raw, SDK ft_raw_force), so the compensation can be redone",
815
  "compensation": "the payload the control box compensated for is in each episode sidecar's metadata.controller_settings.<side>.tcp_load_kg and tcp_load_cog_mm, read at the trial's start",
816
+ "timing": "row k holds the last reading stamped at or before t_k (zero-order hold, fresh within hold_max_age_s); the stamp is when the follower's host read the control box's real-time report packet",
817
  "missing": "rows without a fresh reading hold zeros and paddy.force_torque_valid false; they are not dropped. An arm whose rig declares the sensor but recorded no readings has the columns, all zeros and invalid",
818
+ "rate_hz_nominal": "about 100 (the real-time report); 5 when the follower fell back to the rich report",
819
  "measured_median_hz": {
820
  "right": 99.5
821
  },
 
868
  "action": {
869
  "feature": "action",
870
  "space": "joints",
871
+ "meaning": "per arm the joint command, zero-order held, then the gripper command: the quantity a policy that drives the arms through target_joints outputs. The TCP target the teleop controller published is paddy.target_pose.<side>; the pose the follower sent is paddy.command_sent.<side>",
872
  "joints": "absolute positions of the URDF joints joint1..jointN, base to wrist, in observation.state's order; not a delta, not a TCP pose",
873
  "joint_unit": "rad",
874
  "joints_source": {
 
877
  },
878
  "joints_sources": {
879
  "command_sent": "the joint angles the follower sent (command_sent joints_rad: a set_servo_angle call's target or, in servo mode, a set_servo_angle_j setpoint; the xArm7 joint path, GELLO and policy arms), held with action_hold_s; timing.command_methods says which",
880
+ "target_joints": "a leader arm's recorded joint targets (target_joints, GELLO), held with action_hold_s",
881
  "next_state": "the next row's measured joints (the last row its own): the usual stand-in for a position-controlled arm's joint command when none was recorded (a mode-7 arm, older bags)"
882
  },
883
  "joints_source_per_episode": "conversion.action_source.<side> of meta/paddy/episodes/<trial_id>.json",
 
938
  }
939
  },
940
  "already_scaled": true,
941
+ "scaling_note": "action is in robot units after the controller's motion scaling and rate limit; paddy.scaling is the human-input gain in force (informational, not a policy input)"
942
  },
943
  "timing": {
944
+ "row_time": "t_k = t0 + k / fps (t0 = the latest first stamp over the arms' joint states and the cameras' color frames; paused spans removed); timestamp = k / fps; paddy.utc_s and the sidecar's conversion.t_start_s give the absolute time",
945
  "rule": "row k holds the last joint command stamped at or before t_k (zero-order hold, no shift): the command issued at t_k; with joints_source next_state the measured joints of row k+1, the position the arm reached next",
946
  "stamp_meaning": "command_sent: header.stamp of the command_sent message, when the follower's motion call (set_servo_angle, or set_servo_angle_j in servo mode) returned; target_joints: header.stamp of the leader's target; next_state: the next row's slot time",
947
  "execution": "the follower validates each target and commands the arm from it; how depends on the arm's control mode, named per command by command_sent.method (per arm command_methods, per episode the sidecar's conversion.streams.command_sent.<side>.methods) and described in execution_by_method. The issue time of each command is header.stamp of the command_sent stream",
 
954
  ]
955
  },
956
  "execution_by_method": {
957
+ "set_servo_angle_j": "servo mode (xArm mode 1): the follower takes each accepted target as joints (its own solution of a TCP target, or a leader's joint target), steps a joint setpoint toward the newest one on a fixed tick (rates.control.servo_rate_hz), each joint inside the stream's speed and acceleration caps, and sends every setpoint with set_servo_angle_j; the control box executes the last setpoint it received and plans nothing. command_sent has one entry per setpoint: joints_rad is the setpoint, the pose and target_stamp are those of the target it steps toward"
958
  },
959
  "action_hold_s": 0.2,
960
  "valid_mask": "paddy.action_valid",
 
1000
  "left": "urdf_fk",
1001
  "right": "urdf_fk"
1002
  },
1003
+ "sources": "urdf_fk: the recorded URDF's base_T_tcp at the recorded joint states (100 Hz), the kinematics the wrist cameras' paddy.camera_pose and the teleop targets use; reported: the pose the arm's control box reports (about 50 Hz, its own kinematics, which can differ from the URDF's by several mm: sidecar conversion.kinematics.<side>.tcp_vs_urdf_mm); mixed: episodes differ, see each sidecar's conversion.tcp_pose.<side>",
1004
+ "stamps": "urdf_fk: the joint states' stamps. reported: the control box reports the pose ahead of the arm; each sample is restamped by that lead, measured per episode against the URDF's TCP at the recorded joints, else the rig profile's actual_pose_lead_ms: sidecar conversion.eef_timing.<side>"
1005
  },
1006
  "eef_state": {
1007
  "feature": "observation.eef_state",
1008
+ "meaning": "every arm's measured TCP pose (observation.eef_pose.<side>, sign-continuous) and gripper opening (observation.state's) in one vector, the measured counterpart of a tcp action: the state of an end-effector policy",
1009
  "layout_per_arm": [
1010
  "x",
1011
  "y",
 
1024
  },
1025
  "kinematics": {
1026
  "sidecar": "conversion.kinematics.<side> of meta/paddy/episodes/<trial_id>.json",
1027
+ "tcp_vs_urdf": "tcp_vs_urdf_mm and tcp_vs_urdf_deg (median, max): the URDF's base_T_tcp at the recorded joints against the recorded TCP pose",
1028
+ "dh": "dh: the arm control box's own DH table (metadata.json controller_settings.<side>.dh_params: 28 values, four per joint, mm and rad) as identified against the recorded TCP pose: layout (the name of each of the four values per joint, in the table's order), convention (modified = Craig, or standard), tcp_vs_dh_mm and tcp_vs_dh_deg (median, max), samples, fit, and dh_vs_urdf_mm (the table's flange against the URDF's at the same joints); with fit true, forward kinematics of the table read that way is the arm's calibrated kinematics. {note} when the trial has no table"
1029
  },
1030
  "cameras": {
1031
  "file": "meta/cameras.json",
1032
+ "role": "where the camera is mounted, as the rig profile declares it: overhead (fixed, looking down at the workspace), scene (fixed, another viewpoint) or wrist (moves with an arm; which arm is mount.side); null = not declared",
1033
  "primary": "the file's top-level primary names the rig's one primary camera (the view to look at first), and that camera's entry has primary: true; null = not declared",
1034
  "base_T_cam": "4x4 pose of the camera's color optical frame (z forward, x right, y down) in each arm base frame, metres; extrinsics.<base>.source says where it came from",
1035
  "sources": {
1036
  "tf_static": "the nominal pose: the rig's URDF and the camera mounts in the rig profile (CAD or tape), as published on /tf_static",
1037
  "tf_snapshot": "the nominal pose from the first /tf snapshot at episode start (valid: episode_start)",
1038
+ "urdf_fk": "the nominal pose of a wrist camera by forward kinematics of the recorded URDF and joint states at t_s, the episode's first slot (valid: episode_start)",
1039
+ "measured": "the rig profile's measured calibration: calibration.frame_T_cam chained to the base through via (identity, tf_static, urdf, urdf_fk at t_s for a wrist camera, or tf_snapshot); the nominal entry is kept as nominal {base_T_cam, source}"
1040
  },
1041
  "measured_vs_nominal": "measured_vs_nominal_mm / _deg: distance of the origins and angle of the rotation between the measured and the nominal base_T_cam at the episode's first slot",
1042
+ "intrinsics": "K (row-major 3x3) and D (plumb_bob k1 k2 p1 p2 k3) describe the raw color image as recorded (not undistorted). intrinsics.source camera_info: the camera's own CameraInfo (a RealSense's factory values; a D435 color stream reports D all zeros). measured: the rig profile's measured intrinsics (method, calibrated_at, error_px, samples, notes, from), the CameraInfo values kept as factory_intrinsics {K, D, distortion_model}. The pose of a fixed camera measured after its intrinsics were is solved with them",
1043
  "calibration": "per camera: the measured calibration (frame, xyz_mm, rpy_deg URDF fixed-axis, frame_T_cam, method, calibrated_at, error_mm, error_px, samples, notes, from: rig_snapshot | rig_option), null when the rig profile has none. TF on the rig stays nominal",
1044
+ "image_rotation": "image_rotation_deg per camera: 0, or 180 for a camera mounted upside down whose pictures the dataset turns. Turned, the color and depth frames are the recorded ones turned by 180 degrees and every value that describes them is of the turned image: K (cx' = W - 1 - cx, cy' = H - 1 - cy), D (p1 and p2 negated), base_T_cam, mount link_T_cam and tcp_T_cam, calibration.frame_T_cam, the depth K, D and color_T_depth, and paddy.camera_pose.<camera>: the camera's own optical frame (frame_id) turned about its z axis, X_T_turned = X_T_cam @ diag(-1, -1, 1, 1). calibration.xyz_mm and rpy_deg stay the rig profile's record of the camera's own optical frame. A policy trained on it needs the live frames turned the same way (paddy's policy server does it from this file)",
1045
+ "per_row_pose": "paddy.camera_pose.<camera> of a wrist camera; source measured_fk (forward kinematics to the calibration frame times its measured pose), urdf_fk or eef; time row (at the joint sample observation.state holds, so it matches observation.state through the URDF; the image is paddy.frame_age.<camera> older) or frame (at the shown frame's stamp)"
1046
  },
1047
  "subtasks": {
1048
  "feature": "paddy.subtask_index",
1049
+ "meaning": "the operator-marked subtask open at the row (metadata.json subtasks): its index, -1 where none is open. Spans are half-open [start, end) in header.stamp POSIX seconds; paused spans are out of the grid, so a subtask with a pause inside has no rows for the paused time",
1050
  "names": "the episode sidecar's conversion.subtasks: per index the plain-language instruction, the outcome (success, failure or unknown), start_s, end_s (null: still open at the end of the recording) and the rows [start_row, end_row) it covers (null when none)",
1051
  "not_a_policy_input": true
1052
  },
 
1111
  "frames": {
1112
  "right": "Right_link_eef"
1113
  },
1114
+ "frame_note": "the flange (URDF link_eef), which the six-axis sensor is mounted on. Checked on the rig (validation 1.26, 2026-11-07): a press along each flange axis reads on that axis with the sign of the force applied to the tool (a push toward flange +x reads +fx); a load fit put the sensor axes within 1.4 deg of the flange's",
1115
  "reading": "observation.force_torque.<side>: the control box's reading after its load and offset compensation (/paddy/<side>_arm/ft_wrench, SDK ft_ext_force): near zero, up to a small offset, with nothing touching the tool when the payload set on the control box matches the tool. paddy.force_torque_raw.<side>: the unprocessed reading (/paddy/<side>_arm/ft_wrench_raw, SDK ft_raw_force), so the compensation can be redone",
1116
  "compensation": "the payload the control box compensated for is in each episode sidecar's metadata.controller_settings.<side>.tcp_load_kg and tcp_load_cog_mm, read at the trial's start",
1117
+ "timing": "row k holds the last reading stamped at or before t_k (zero-order hold, fresh within hold_max_age_s); the stamp is when the follower's host read the control box's real-time report packet",
1118
  "missing": "rows without a fresh reading hold zeros and paddy.force_torque_valid false; they are not dropped. An arm whose rig declares the sensor but recorded no readings has the columns, all zeros and invalid",
1119
+ "rate_hz_nominal": "about 100 (the real-time report); 5 when the follower fell back to the rich report",
1120
  "measured_median_hz": {
1121
  "right": 99.5
1122
  },
 
1169
  "action": {
1170
  "feature": "action",
1171
  "space": "joints",
1172
+ "meaning": "per arm the joint command, zero-order held, then the gripper command: the quantity a policy that drives the arms through target_joints outputs. The TCP target the teleop controller published is paddy.target_pose.<side>; the pose the follower sent is paddy.command_sent.<side>",
1173
  "joints": "absolute positions of the URDF joints joint1..jointN, base to wrist, in observation.state's order; not a delta, not a TCP pose",
1174
  "joint_unit": "rad",
1175
  "joints_source": {
 
1178
  },
1179
  "joints_sources": {
1180
  "command_sent": "the joint angles the follower sent (command_sent joints_rad: a set_servo_angle call's target or, in servo mode, a set_servo_angle_j setpoint; the xArm7 joint path, GELLO and policy arms), held with action_hold_s; timing.command_methods says which",
1181
+ "target_joints": "a leader arm's recorded joint targets (target_joints, GELLO), held with action_hold_s",
1182
  "next_state": "the next row's measured joints (the last row its own): the usual stand-in for a position-controlled arm's joint command when none was recorded (a mode-7 arm, older bags)"
1183
  },
1184
  "joints_source_per_episode": "conversion.action_source.<side> of meta/paddy/episodes/<trial_id>.json",
 
1239
  }
1240
  },
1241
  "already_scaled": true,
1242
+ "scaling_note": "action is in robot units after the controller's motion scaling and rate limit; paddy.scaling is the human-input gain in force (informational, not a policy input)"
1243
  },
1244
  "timing": {
1245
+ "row_time": "t_k = t0 + k / fps (t0 = the latest first stamp over the arms' joint states and the cameras' color frames; paused spans removed); timestamp = k / fps; paddy.utc_s and the sidecar's conversion.t_start_s give the absolute time",
1246
  "rule": "row k holds the last joint command stamped at or before t_k (zero-order hold, no shift): the command issued at t_k; with joints_source next_state the measured joints of row k+1, the position the arm reached next",
1247
  "stamp_meaning": "command_sent: header.stamp of the command_sent message, when the follower's motion call (set_servo_angle, or set_servo_angle_j in servo mode) returned; target_joints: header.stamp of the leader's target; next_state: the next row's slot time",
1248
  "execution": "the follower validates each target and commands the arm from it; how depends on the arm's control mode, named per command by command_sent.method (per arm command_methods, per episode the sidecar's conversion.streams.command_sent.<side>.methods) and described in execution_by_method. The issue time of each command is header.stamp of the command_sent stream",
 
1255
  ]
1256
  },
1257
  "execution_by_method": {
1258
+ "set_servo_angle_j": "servo mode (xArm mode 1): the follower takes each accepted target as joints (its own solution of a TCP target, or a leader's joint target), steps a joint setpoint toward the newest one on a fixed tick (rates.control.servo_rate_hz), each joint inside the stream's speed and acceleration caps, and sends every setpoint with set_servo_angle_j; the control box executes the last setpoint it received and plans nothing. command_sent has one entry per setpoint: joints_rad is the setpoint, the pose and target_stamp are those of the target it steps toward"
1259
  },
1260
  "action_hold_s": 0.2,
1261
  "valid_mask": "paddy.action_valid",
 
1301
  "left": "urdf_fk",
1302
  "right": "urdf_fk"
1303
  },
1304
+ "sources": "urdf_fk: the recorded URDF's base_T_tcp at the recorded joint states (100 Hz), the kinematics the wrist cameras' paddy.camera_pose and the teleop targets use; reported: the pose the arm's control box reports (about 50 Hz, its own kinematics, which can differ from the URDF's by several mm: sidecar conversion.kinematics.<side>.tcp_vs_urdf_mm); mixed: episodes differ, see each sidecar's conversion.tcp_pose.<side>",
1305
+ "stamps": "urdf_fk: the joint states' stamps. reported: the control box reports the pose ahead of the arm; each sample is restamped by that lead, measured per episode against the URDF's TCP at the recorded joints, else the rig profile's actual_pose_lead_ms: sidecar conversion.eef_timing.<side>"
1306
  },
1307
  "eef_state": {
1308
  "feature": "observation.eef_state",
1309
+ "meaning": "every arm's measured TCP pose (observation.eef_pose.<side>, sign-continuous) and gripper opening (observation.state's) in one vector, the measured counterpart of a tcp action: the state of an end-effector policy",
1310
  "layout_per_arm": [
1311
  "x",
1312
  "y",
 
1325
  },
1326
  "kinematics": {
1327
  "sidecar": "conversion.kinematics.<side> of meta/paddy/episodes/<trial_id>.json",
1328
+ "tcp_vs_urdf": "tcp_vs_urdf_mm and tcp_vs_urdf_deg (median, max): the URDF's base_T_tcp at the recorded joints against the recorded TCP pose",
1329
+ "dh": "dh: the arm control box's own DH table (metadata.json controller_settings.<side>.dh_params: 28 values, four per joint, mm and rad) as identified against the recorded TCP pose: layout (the name of each of the four values per joint, in the table's order), convention (modified = Craig, or standard), tcp_vs_dh_mm and tcp_vs_dh_deg (median, max), samples, fit, and dh_vs_urdf_mm (the table's flange against the URDF's at the same joints); with fit true, forward kinematics of the table read that way is the arm's calibrated kinematics. {note} when the trial has no table"
1330
  },
1331
  "cameras": {
1332
  "file": "meta/cameras.json",
1333
+ "role": "where the camera is mounted, as the rig profile declares it: overhead (fixed, looking down at the workspace), scene (fixed, another viewpoint) or wrist (moves with an arm; which arm is mount.side); null = not declared",
1334
  "primary": "the file's top-level primary names the rig's one primary camera (the view to look at first), and that camera's entry has primary: true; null = not declared",
1335
  "base_T_cam": "4x4 pose of the camera's color optical frame (z forward, x right, y down) in each arm base frame, metres; extrinsics.<base>.source says where it came from",
1336
  "sources": {
1337
  "tf_static": "the nominal pose: the rig's URDF and the camera mounts in the rig profile (CAD or tape), as published on /tf_static",
1338
  "tf_snapshot": "the nominal pose from the first /tf snapshot at episode start (valid: episode_start)",
1339
+ "urdf_fk": "the nominal pose of a wrist camera by forward kinematics of the recorded URDF and joint states at t_s, the episode's first slot (valid: episode_start)",
1340
+ "measured": "the rig profile's measured calibration: calibration.frame_T_cam chained to the base through via (identity, tf_static, urdf, urdf_fk at t_s for a wrist camera, or tf_snapshot); the nominal entry is kept as nominal {base_T_cam, source}"
1341
  },
1342
  "measured_vs_nominal": "measured_vs_nominal_mm / _deg: distance of the origins and angle of the rotation between the measured and the nominal base_T_cam at the episode's first slot",
1343
+ "intrinsics": "K (row-major 3x3) and D (plumb_bob k1 k2 p1 p2 k3) describe the raw color image as recorded (not undistorted). intrinsics.source camera_info: the camera's own CameraInfo (a RealSense's factory values; a D435 color stream reports D all zeros). measured: the rig profile's measured intrinsics (method, calibrated_at, error_px, samples, notes, from), the CameraInfo values kept as factory_intrinsics {K, D, distortion_model}. The pose of a fixed camera measured after its intrinsics were is solved with them",
1344
  "calibration": "per camera: the measured calibration (frame, xyz_mm, rpy_deg URDF fixed-axis, frame_T_cam, method, calibrated_at, error_mm, error_px, samples, notes, from: rig_snapshot | rig_option), null when the rig profile has none. TF on the rig stays nominal",
1345
+ "image_rotation": "image_rotation_deg per camera: 0, or 180 for a camera mounted upside down whose pictures the dataset turns. Turned, the color and depth frames are the recorded ones turned by 180 degrees and every value that describes them is of the turned image: K (cx' = W - 1 - cx, cy' = H - 1 - cy), D (p1 and p2 negated), base_T_cam, mount link_T_cam and tcp_T_cam, calibration.frame_T_cam, the depth K, D and color_T_depth, and paddy.camera_pose.<camera>: the camera's own optical frame (frame_id) turned about its z axis, X_T_turned = X_T_cam @ diag(-1, -1, 1, 1). calibration.xyz_mm and rpy_deg stay the rig profile's record of the camera's own optical frame. A policy trained on it needs the live frames turned the same way (paddy's policy server does it from this file)",
1346
+ "per_row_pose": "paddy.camera_pose.<camera> of a wrist camera; source measured_fk (forward kinematics to the calibration frame times its measured pose), urdf_fk or eef; time row (at the joint sample observation.state holds, so it matches observation.state through the URDF; the image is paddy.frame_age.<camera> older) or frame (at the shown frame's stamp)"
1347
  },
1348
  "subtasks": {
1349
  "feature": "paddy.subtask_index",
1350
+ "meaning": "the operator-marked subtask open at the row (metadata.json subtasks): its index, -1 where none is open. Spans are half-open [start, end) in header.stamp POSIX seconds; paused spans are out of the grid, so a subtask with a pause inside has no rows for the paused time",
1351
  "names": "the episode sidecar's conversion.subtasks: per index the plain-language instruction, the outcome (success, failure or unknown), start_s, end_s (null: still open at the end of the recording) and the rows [start_row, end_row) it covers (null when none)",
1352
  "not_a_policy_input": true
1353
  },
 
1412
  "frames": {
1413
  "right": "Right_link_eef"
1414
  },
1415
+ "frame_note": "the flange (URDF link_eef), which the six-axis sensor is mounted on. Checked on the rig (validation 1.26, 2026-11-07): a press along each flange axis reads on that axis with the sign of the force applied to the tool (a push toward flange +x reads +fx); a load fit put the sensor axes within 1.4 deg of the flange's",
1416
  "reading": "observation.force_torque.<side>: the control box's reading after its load and offset compensation (/paddy/<side>_arm/ft_wrench, SDK ft_ext_force): near zero, up to a small offset, with nothing touching the tool when the payload set on the control box matches the tool. paddy.force_torque_raw.<side>: the unprocessed reading (/paddy/<side>_arm/ft_wrench_raw, SDK ft_raw_force), so the compensation can be redone",
1417
  "compensation": "the payload the control box compensated for is in each episode sidecar's metadata.controller_settings.<side>.tcp_load_kg and tcp_load_cog_mm, read at the trial's start",
1418
+ "timing": "row k holds the last reading stamped at or before t_k (zero-order hold, fresh within hold_max_age_s); the stamp is when the follower's host read the control box's real-time report packet",
1419
  "missing": "rows without a fresh reading hold zeros and paddy.force_torque_valid false; they are not dropped. An arm whose rig declares the sensor but recorded no readings has the columns, all zeros and invalid",
1420
+ "rate_hz_nominal": "about 100 (the real-time report); 5 when the follower fell back to the rich report",
1421
  "measured_median_hz": {
1422
  "right": 99.5
1423
  },
 
1470
  "action": {
1471
  "feature": "action",
1472
  "space": "joints",
1473
+ "meaning": "per arm the joint command, zero-order held, then the gripper command: the quantity a policy that drives the arms through target_joints outputs. The TCP target the teleop controller published is paddy.target_pose.<side>; the pose the follower sent is paddy.command_sent.<side>",
1474
  "joints": "absolute positions of the URDF joints joint1..jointN, base to wrist, in observation.state's order; not a delta, not a TCP pose",
1475
  "joint_unit": "rad",
1476
  "joints_source": {
 
1479
  },
1480
  "joints_sources": {
1481
  "command_sent": "the joint angles the follower sent (command_sent joints_rad: a set_servo_angle call's target or, in servo mode, a set_servo_angle_j setpoint; the xArm7 joint path, GELLO and policy arms), held with action_hold_s; timing.command_methods says which",
1482
+ "target_joints": "a leader arm's recorded joint targets (target_joints, GELLO), held with action_hold_s",
1483
  "next_state": "the next row's measured joints (the last row its own): the usual stand-in for a position-controlled arm's joint command when none was recorded (a mode-7 arm, older bags)"
1484
  },
1485
  "joints_source_per_episode": "conversion.action_source.<side> of meta/paddy/episodes/<trial_id>.json",
 
1540
  }
1541
  },
1542
  "already_scaled": true,
1543
+ "scaling_note": "action is in robot units after the controller's motion scaling and rate limit; paddy.scaling is the human-input gain in force (informational, not a policy input)"
1544
  },
1545
  "timing": {
1546
+ "row_time": "t_k = t0 + k / fps (t0 = the latest first stamp over the arms' joint states and the cameras' color frames; paused spans removed); timestamp = k / fps; paddy.utc_s and the sidecar's conversion.t_start_s give the absolute time",
1547
  "rule": "row k holds the last joint command stamped at or before t_k (zero-order hold, no shift): the command issued at t_k; with joints_source next_state the measured joints of row k+1, the position the arm reached next",
1548
  "stamp_meaning": "command_sent: header.stamp of the command_sent message, when the follower's motion call (set_servo_angle, or set_servo_angle_j in servo mode) returned; target_joints: header.stamp of the leader's target; next_state: the next row's slot time",
1549
  "execution": "the follower validates each target and commands the arm from it; how depends on the arm's control mode, named per command by command_sent.method (per arm command_methods, per episode the sidecar's conversion.streams.command_sent.<side>.methods) and described in execution_by_method. The issue time of each command is header.stamp of the command_sent stream",
 
1556
  ]
1557
  },
1558
  "execution_by_method": {
1559
+ "set_servo_angle_j": "servo mode (xArm mode 1): the follower takes each accepted target as joints (its own solution of a TCP target, or a leader's joint target), steps a joint setpoint toward the newest one on a fixed tick (rates.control.servo_rate_hz), each joint inside the stream's speed and acceleration caps, and sends every setpoint with set_servo_angle_j; the control box executes the last setpoint it received and plans nothing. command_sent has one entry per setpoint: joints_rad is the setpoint, the pose and target_stamp are those of the target it steps toward"
1560
  },
1561
  "action_hold_s": 0.2,
1562
  "valid_mask": "paddy.action_valid",
 
1602
  "left": "urdf_fk",
1603
  "right": "urdf_fk"
1604
  },
1605
+ "sources": "urdf_fk: the recorded URDF's base_T_tcp at the recorded joint states (100 Hz), the kinematics the wrist cameras' paddy.camera_pose and the teleop targets use; reported: the pose the arm's control box reports (about 50 Hz, its own kinematics, which can differ from the URDF's by several mm: sidecar conversion.kinematics.<side>.tcp_vs_urdf_mm); mixed: episodes differ, see each sidecar's conversion.tcp_pose.<side>",
1606
+ "stamps": "urdf_fk: the joint states' stamps. reported: the control box reports the pose ahead of the arm; each sample is restamped by that lead, measured per episode against the URDF's TCP at the recorded joints, else the rig profile's actual_pose_lead_ms: sidecar conversion.eef_timing.<side>"
1607
  },
1608
  "eef_state": {
1609
  "feature": "observation.eef_state",
1610
+ "meaning": "every arm's measured TCP pose (observation.eef_pose.<side>, sign-continuous) and gripper opening (observation.state's) in one vector, the measured counterpart of a tcp action: the state of an end-effector policy",
1611
  "layout_per_arm": [
1612
  "x",
1613
  "y",
 
1626
  },
1627
  "kinematics": {
1628
  "sidecar": "conversion.kinematics.<side> of meta/paddy/episodes/<trial_id>.json",
1629
+ "tcp_vs_urdf": "tcp_vs_urdf_mm and tcp_vs_urdf_deg (median, max): the URDF's base_T_tcp at the recorded joints against the recorded TCP pose",
1630
+ "dh": "dh: the arm control box's own DH table (metadata.json controller_settings.<side>.dh_params: 28 values, four per joint, mm and rad) as identified against the recorded TCP pose: layout (the name of each of the four values per joint, in the table's order), convention (modified = Craig, or standard), tcp_vs_dh_mm and tcp_vs_dh_deg (median, max), samples, fit, and dh_vs_urdf_mm (the table's flange against the URDF's at the same joints); with fit true, forward kinematics of the table read that way is the arm's calibrated kinematics. {note} when the trial has no table"
1631
  },
1632
  "cameras": {
1633
  "file": "meta/cameras.json",
1634
+ "role": "where the camera is mounted, as the rig profile declares it: overhead (fixed, looking down at the workspace), scene (fixed, another viewpoint) or wrist (moves with an arm; which arm is mount.side); null = not declared",
1635
  "primary": "the file's top-level primary names the rig's one primary camera (the view to look at first), and that camera's entry has primary: true; null = not declared",
1636
  "base_T_cam": "4x4 pose of the camera's color optical frame (z forward, x right, y down) in each arm base frame, metres; extrinsics.<base>.source says where it came from",
1637
  "sources": {
1638
  "tf_static": "the nominal pose: the rig's URDF and the camera mounts in the rig profile (CAD or tape), as published on /tf_static",
1639
  "tf_snapshot": "the nominal pose from the first /tf snapshot at episode start (valid: episode_start)",
1640
+ "urdf_fk": "the nominal pose of a wrist camera by forward kinematics of the recorded URDF and joint states at t_s, the episode's first slot (valid: episode_start)",
1641
+ "measured": "the rig profile's measured calibration: calibration.frame_T_cam chained to the base through via (identity, tf_static, urdf, urdf_fk at t_s for a wrist camera, or tf_snapshot); the nominal entry is kept as nominal {base_T_cam, source}"
1642
  },
1643
  "measured_vs_nominal": "measured_vs_nominal_mm / _deg: distance of the origins and angle of the rotation between the measured and the nominal base_T_cam at the episode's first slot",
1644
+ "intrinsics": "K (row-major 3x3) and D (plumb_bob k1 k2 p1 p2 k3) describe the raw color image as recorded (not undistorted). intrinsics.source camera_info: the camera's own CameraInfo (a RealSense's factory values; a D435 color stream reports D all zeros). measured: the rig profile's measured intrinsics (method, calibrated_at, error_px, samples, notes, from), the CameraInfo values kept as factory_intrinsics {K, D, distortion_model}. The pose of a fixed camera measured after its intrinsics were is solved with them",
1645
  "calibration": "per camera: the measured calibration (frame, xyz_mm, rpy_deg URDF fixed-axis, frame_T_cam, method, calibrated_at, error_mm, error_px, samples, notes, from: rig_snapshot | rig_option), null when the rig profile has none. TF on the rig stays nominal",
1646
+ "image_rotation": "image_rotation_deg per camera: 0, or 180 for a camera mounted upside down whose pictures the dataset turns. Turned, the color and depth frames are the recorded ones turned by 180 degrees and every value that describes them is of the turned image: K (cx' = W - 1 - cx, cy' = H - 1 - cy), D (p1 and p2 negated), base_T_cam, mount link_T_cam and tcp_T_cam, calibration.frame_T_cam, the depth K, D and color_T_depth, and paddy.camera_pose.<camera>: the camera's own optical frame (frame_id) turned about its z axis, X_T_turned = X_T_cam @ diag(-1, -1, 1, 1). calibration.xyz_mm and rpy_deg stay the rig profile's record of the camera's own optical frame. A policy trained on it needs the live frames turned the same way (paddy's policy server does it from this file)",
1647
+ "per_row_pose": "paddy.camera_pose.<camera> of a wrist camera; source measured_fk (forward kinematics to the calibration frame times its measured pose), urdf_fk or eef; time row (at the joint sample observation.state holds, so it matches observation.state through the URDF; the image is paddy.frame_age.<camera> older) or frame (at the shown frame's stamp)"
1648
  },
1649
  "subtasks": {
1650
  "feature": "paddy.subtask_index",
1651
+ "meaning": "the operator-marked subtask open at the row (metadata.json subtasks): its index, -1 where none is open. Spans are half-open [start, end) in header.stamp POSIX seconds; paused spans are out of the grid, so a subtask with a pause inside has no rows for the paused time",
1652
  "names": "the episode sidecar's conversion.subtasks: per index the plain-language instruction, the outcome (success, failure or unknown), start_s, end_s (null: still open at the end of the recording) and the rows [start_row, end_row) it covers (null when none)",
1653
  "not_a_policy_input": true
1654
  },
 
1713
  "frames": {
1714
  "right": "Right_link_eef"
1715
  },
1716
+ "frame_note": "the flange (URDF link_eef), which the six-axis sensor is mounted on. Checked on the rig (validation 1.26, 2026-11-07): a press along each flange axis reads on that axis with the sign of the force applied to the tool (a push toward flange +x reads +fx); a load fit put the sensor axes within 1.4 deg of the flange's",
1717
  "reading": "observation.force_torque.<side>: the control box's reading after its load and offset compensation (/paddy/<side>_arm/ft_wrench, SDK ft_ext_force): near zero, up to a small offset, with nothing touching the tool when the payload set on the control box matches the tool. paddy.force_torque_raw.<side>: the unprocessed reading (/paddy/<side>_arm/ft_wrench_raw, SDK ft_raw_force), so the compensation can be redone",
1718
  "compensation": "the payload the control box compensated for is in each episode sidecar's metadata.controller_settings.<side>.tcp_load_kg and tcp_load_cog_mm, read at the trial's start",
1719
+ "timing": "row k holds the last reading stamped at or before t_k (zero-order hold, fresh within hold_max_age_s); the stamp is when the follower's host read the control box's real-time report packet",
1720
  "missing": "rows without a fresh reading hold zeros and paddy.force_torque_valid false; they are not dropped. An arm whose rig declares the sensor but recorded no readings has the columns, all zeros and invalid",
1721
+ "rate_hz_nominal": "about 100 (the real-time report); 5 when the follower fell back to the rich report",
1722
  "measured_median_hz": {
1723
  "right": 99.5
1724
  },
meta/paddy/robot_description.urdf CHANGED
@@ -174,7 +174,7 @@
174
  </joint>
175
  <link name="bottom_camera_swing_link">
176
  <visual>
177
- <!-- the D405 swing plate (ADR-0208), as the IN rig's top D405: link origin on the
178
  pivot hole, 0.026 above the plate's camera face -->
179
  <origin rpy="0 0 0" xyz="0 0 -0.026"/>
180
  <geometry>
@@ -999,7 +999,7 @@
999
  <maxEffort>10.0</maxEffort>
1000
  </plugin>
1001
  </gazebo>
1002
- <!-- paddy (ADR-0127): paddy's D405 bracket (d405_camerastand_10092026.step, meshed with
1003
  its origin at the centre of the arm-side face, z towards the gripper, the tab along
1004
  +x, mm) right under the gripper: its tool-side face is the gripper's seat,
1005
  gripper_mount_z_1 out from the flange, so its arm-side face is 3 mm closer
@@ -1021,7 +1021,7 @@
1021
  <child link="Left_d405_bracket_link"/>
1022
  <origin rpy="0 0 0" xyz="0 0 0.005999999999999999"/>
1023
  </joint>
1024
- <!-- The D405 as on the wing rig (ADR-0071): <prefix>camera_attach_link at
1025
  wrist_camera_xyz/rpy_1 in link_eef, and the macro's own bottom-screw-to-link offset
1026
  negated so <prefix>camera_link lands exactly on it. -->
1027
  <link name="Left_camera_attach_link"/>
@@ -1818,7 +1818,7 @@
1818
  <maxEffort>10.0</maxEffort>
1819
  </plugin>
1820
  </gazebo>
1821
- <!-- paddy (ADR-0127): paddy's D405 bracket (d405_camerastand_10092026.step, meshed with
1822
  its origin at the centre of the arm-side face, z towards the gripper, the tab along
1823
  +x, mm) right under the gripper: its tool-side face is the gripper's seat,
1824
  gripper_mount_z_2 out from the flange, so its arm-side face is 3 mm closer
@@ -1840,7 +1840,7 @@
1840
  <child link="Right_d405_bracket_link"/>
1841
  <origin rpy="0 0 0" xyz="0 0 0.0643"/>
1842
  </joint>
1843
- <!-- The D405 as on the wing rig (ADR-0071): <prefix>camera_attach_link at
1844
  wrist_camera_xyz/rpy_2 in link_eef, and the macro's own bottom-screw-to-link offset
1845
  negated so <prefix>camera_link lands exactly on it. -->
1846
  <link name="Right_camera_attach_link"/>
 
174
  </joint>
175
  <link name="bottom_camera_swing_link">
176
  <visual>
177
+ <!-- the D405 swing plate, as the IN rig's top D405: link origin on the
178
  pivot hole, 0.026 above the plate's camera face -->
179
  <origin rpy="0 0 0" xyz="0 0 -0.026"/>
180
  <geometry>
 
999
  <maxEffort>10.0</maxEffort>
1000
  </plugin>
1001
  </gazebo>
1002
+ <!-- paddy: paddy's D405 bracket (d405_camerastand_10092026.step, meshed with
1003
  its origin at the centre of the arm-side face, z towards the gripper, the tab along
1004
  +x, mm) right under the gripper: its tool-side face is the gripper's seat,
1005
  gripper_mount_z_1 out from the flange, so its arm-side face is 3 mm closer
 
1021
  <child link="Left_d405_bracket_link"/>
1022
  <origin rpy="0 0 0" xyz="0 0 0.005999999999999999"/>
1023
  </joint>
1024
+ <!-- The D405 as on the wing rig: <prefix>camera_attach_link at
1025
  wrist_camera_xyz/rpy_1 in link_eef, and the macro's own bottom-screw-to-link offset
1026
  negated so <prefix>camera_link lands exactly on it. -->
1027
  <link name="Left_camera_attach_link"/>
 
1818
  <maxEffort>10.0</maxEffort>
1819
  </plugin>
1820
  </gazebo>
1821
+ <!-- paddy: paddy's D405 bracket (d405_camerastand_10092026.step, meshed with
1822
  its origin at the centre of the arm-side face, z towards the gripper, the tab along
1823
  +x, mm) right under the gripper: its tool-side face is the gripper's seat,
1824
  gripper_mount_z_2 out from the flange, so its arm-side face is 3 mm closer
 
1840
  <child link="Right_d405_bracket_link"/>
1841
  <origin rpy="0 0 0" xyz="0 0 0.0643"/>
1842
  </joint>
1843
+ <!-- The D405 as on the wing rig: <prefix>camera_attach_link at
1844
  wrist_camera_xyz/rpy_2 in link_eef, and the macro's own bottom-screw-to-link offset
1845
  negated so <prefix>camera_link lands exactly on it. -->
1846
  <link name="Right_camera_attach_link"/>
meta/paddy/stereo_calibration_camera_top.json CHANGED
@@ -1397,5 +1397,5 @@
1397
  }
1398
  ],
1399
  "calibrated_at": "2026-11-17T16:20:58+00:00",
1400
- "made_by": "paddy calibrate stereo (ADR-0228)"
1401
  }
 
1397
  }
1398
  ],
1399
  "calibrated_at": "2026-11-17T16:20:58+00:00",
1400
+ "made_by": "paddy calibrate stereo"
1401
  }