ChaoyiPan commited on
Commit
d08585b
·
verified ·
1 Parent(s): ecb9a9e

Replace example archives with object-only pipeline materials and remove hand-only records

Browse files
agentic_pipeline/2026-10-06/README.md CHANGED
@@ -33,10 +33,6 @@ Examples that run locally without neural model weights:
33
  uv run -m examples.inspect_motion --dataset-dir example_datasets/video_examples \
34
  --dataset-name video --task whisking --embodiment-type right
35
 
36
- # Inspect the new native EgoVerse hand retarget.
37
- uv run examples/inspect_dataset.py --dataset-dir example_datasets/egoverse \
38
- --dataset-name egoverse_hand --robot-type sharpa --data-type kinematic
39
-
40
  # Replay the published upstream Sharpa whisking optimization.
41
  uv run examples/inspect_dataset.py --dataset-dir example_datasets/video_examples \
42
  --dataset-name do_as_i_do_reference --robot-type sharpa
@@ -45,19 +41,23 @@ uv run examples/inspect_dataset.py --dataset-dir example_datasets/video_examples
45
  Run one viewer at a time, or assign different `--port` values. Each prints its
46
  browser URL. Model inference is unnecessary for these precomputed examples.
47
 
 
 
 
48
  ## Contents and evidence
49
 
50
  - `source.tar.gz`: the complete task source snapshot, Sharpa assets, skills,
51
  documentation, tests, and third-party notices. Keep the Claude skill symlinks.
52
  - `datasets.tar.gz`: `spider/example_datasets/pipeline_verification`,
53
- `video_examples`, and `egoverse`.
54
  - `reports/`: machine-readable checks and quality diagnostics.
55
  - `previews/`: actual MuJoCo renders from GPU scene checks.
56
 
57
- **144 tests and 21 subtests passed.** Final Sharpa GPU stepping/rendering
58
- checks passed. The four-step native hand optimization probe measured 2.84 mm
59
- mean and 4.70 mm maximum hand-site error against its IK reference; it is a
60
- bounded integration check, not full-clip or manipulation success.
 
61
 
62
  The 40 public source clips cover GigaHands, OakInk, OakInk-v2, FAIR-FRE, and
63
  FAIR-MON. All passed full-clip Sharpa IK: **11,626 frames**. Earlier explicit
@@ -70,16 +70,10 @@ The separate `do_as_i_do_reference` dataset contains **1,000 recorded Sharpa
70
  physics frames**, imported from the upstream result after removing warmup.
71
  It is explicitly an upstream optimization, not a new SPIDER optimization.
72
 
73
- EgoVerse includes two separate community samples: an RGB cup-pickup clip and
74
- native hand tracks. The native example preserves **120 frames at 30 Hz** and
75
- retargets all five fingers to Sharpa. It models hand motion only: there is no
76
- object or cloth simulation, and no manipulation-success claim. Source cards,
77
- checksums, licenses, frame conventions, and residuals are included.
78
-
79
  ## Known quality limits
80
 
81
- Fresh whisking physics does **not** pass the strict initial-grasp stability
82
- check. The diagnostic includes a 12-candidate Xhand search with no passing
83
  candidate; the new Sharpa preparation also needs grasp refinement. The guard
84
  remains enabled. Use the labeled upstream result for recorded replay, and
85
  inspect hand–object alignment before trying a new optimization.
@@ -103,8 +97,6 @@ recommendation. Compare full clips and warmed runtimes for quality decisions.
103
  and the Sharpa asset `SOURCE.md`.
104
  - Public pipeline inputs: `retarget/retarget_example`, revision
105
  `8b0cdf2b24cc86d1c859050c5b0317ddbec59a63`; their original dataset terms apply.
106
- - EgoVerse community examples: Apache-2.0 video repack and CC-BY-4.0 native
107
- sample; source cards, exact revisions, attribution, and licenses accompany them.
108
  - The original `pickcup_right/` RGB/depth sequence in this Hugging Face repo is
109
  preserved. The package also contains its staged RGB video and frame receipt;
110
  it does not claim a reconstructed hand/object trajectory for that video.
 
33
  uv run -m examples.inspect_motion --dataset-dir example_datasets/video_examples \
34
  --dataset-name video --task whisking --embodiment-type right
35
 
 
 
 
 
36
  # Replay the published upstream Sharpa whisking optimization.
37
  uv run examples/inspect_dataset.py --dataset-dir example_datasets/video_examples \
38
  --dataset-name do_as_i_do_reference --robot-type sharpa
 
41
  Run one viewer at a time, or assign different `--port` values. Each prints its
42
  browser URL. Model inference is unnecessary for these precomputed examples.
43
 
44
+ The refreshed object-only source archive passed **232 CPU tests and 21 subtests** after extraction.
45
+ Full object-manipulation quality is reported separately for each actual run.
46
+
47
  ## Contents and evidence
48
 
49
  - `source.tar.gz`: the complete task source snapshot, Sharpa assets, skills,
50
  documentation, tests, and third-party notices. Keep the Claude skill symlinks.
51
  - `datasets.tar.gz`: `spider/example_datasets/pipeline_verification`,
52
+ and `video_examples`.
53
  - `reports/`: machine-readable checks and quality diagnostics.
54
  - `previews/`: actual MuJoCo renders from GPU scene checks.
55
 
56
+ Full physics attempts and their videos are published under
57
+ [`2026-10-06-full-retarget`](../2026-10-06-full-retarget/).
58
+ Inspect the source hand/object alignment, actual object motion, grasp retention,
59
+ and quality report for each attempt. Completed execution alone does not establish
60
+ correct retargeting. Failed attempts remain labeled as failures.
61
 
62
  The 40 public source clips cover GigaHands, OakInk, OakInk-v2, FAIR-FRE, and
63
  FAIR-MON. All passed full-clip Sharpa IK: **11,626 frames**. Earlier explicit
 
70
  physics frames**, imported from the upstream result after removing warmup.
71
  It is explicitly an upstream optimization, not a new SPIDER optimization.
72
 
 
 
 
 
 
 
73
  ## Known quality limits
74
 
75
+ The original static-start whisking path does **not** pass the strict initial-grasp stability
76
+ check. A separate full video backend now includes grasp initialization; its measured attempts are linked above. The diagnostic includes a 12-candidate Xhand search with no passing
77
  candidate; the new Sharpa preparation also needs grasp refinement. The guard
78
  remains enabled. Use the labeled upstream result for recorded replay, and
79
  inspect hand–object alignment before trying a new optimization.
 
97
  and the Sharpa asset `SOURCE.md`.
98
  - Public pipeline inputs: `retarget/retarget_example`, revision
99
  `8b0cdf2b24cc86d1c859050c5b0317ddbec59a63`; their original dataset terms apply.
 
 
100
  - The original `pickcup_right/` RGB/depth sequence in this Hugging Face repo is
101
  preserved. The package also contains its staged RGB video and frame receipt;
102
  it does not claim a reconstructed hand/object trajectory for that video.
agentic_pipeline/2026-10-06/SHA256SUMS CHANGED
@@ -1,3 +1,3 @@
1
- 3d657049c26d73de278fb4e037edd7a7f42b2721a2a5262c523b83d1f6718826 source.tar.gz
2
- 65d59800099472d739889dc1ce0621b16a1bfeb8f582f9c9c802ac3b6ad61c8c datasets.tar.gz
3
- 94426166e91a5fad056e4cfde482b5ae28f386891ee49b52dbbefc4636a9dc46 manifest.json
 
1
+ 2724c4dad8b172fec0bbc11f54a636345c4289ca397574d5fec99bf88e91525c source.tar.gz
2
+ 0ae621c2980784ed4309a630639caa0b418cb4196cac1fb07a3d73a3852e2229 datasets.tar.gz
3
+ 18eeb9a2bf7a145a9084166fa095f24a299074cdfd46ee7a13e5aaa5b5775cec manifest.json
agentic_pipeline/2026-10-06/datasets.tar.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:65d59800099472d739889dc1ce0621b16a1bfeb8f582f9c9c802ac3b6ad61c8c
3
- size 740945873
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0ae621c2980784ed4309a630639caa0b418cb4196cac1fb07a3d73a3852e2229
3
+ size 706172130
agentic_pipeline/2026-10-06/manifest.json CHANGED
@@ -1,28 +1,27 @@
1
  {
2
  "schema_version": 1,
3
- "created_at": "2026-10-06T21:02:07.339568+00:00",
4
  "source_commit": "12c80e2c27e2d9af1bc6c051537047d6bd6640d5",
5
- "source_has_task_changes": false,
6
  "archives": {
7
  "source.tar.gz": {
8
- "sha256": "3d657049c26d73de278fb4e037edd7a7f42b2721a2a5262c523b83d1f6718826",
9
- "bytes": 183064835,
10
- "files": 857
11
  },
12
  "datasets.tar.gz": {
13
- "sha256": "65d59800099472d739889dc1ce0621b16a1bfeb8f582f9c9c802ac3b6ad61c8c",
14
- "bytes": 740945873,
15
- "files": 2893
16
  }
17
  },
18
  "data_directories": [
19
  "example_datasets/pipeline_verification",
20
- "example_datasets/video_examples",
21
- "example_datasets/egoverse"
22
  ],
23
  "notes": [
24
  "Source archive includes current task changes and both agents' skills.",
25
  "Neural model weights and MANO assets are not included; precomputed reconstruction is included.",
26
  "Execution receipts retain original paths as provenance; use a fresh output directory for new runs."
27
  ]
28
- }
 
1
  {
2
  "schema_version": 1,
3
+ "created_at": "2026-10-06T22:26:07.385049+00:00",
4
  "source_commit": "12c80e2c27e2d9af1bc6c051537047d6bd6640d5",
5
+ "source_has_task_changes": true,
6
  "archives": {
7
  "source.tar.gz": {
8
+ "sha256": "2724c4dad8b172fec0bbc11f54a636345c4289ca397574d5fec99bf88e91525c",
9
+ "bytes": 183276422,
10
+ "files": 911
11
  },
12
  "datasets.tar.gz": {
13
+ "sha256": "0ae621c2980784ed4309a630639caa0b418cb4196cac1fb07a3d73a3852e2229",
14
+ "bytes": 706172130,
15
+ "files": 2662
16
  }
17
  },
18
  "data_directories": [
19
  "example_datasets/pipeline_verification",
20
+ "example_datasets/video_examples"
 
21
  ],
22
  "notes": [
23
  "Source archive includes current task changes and both agents' skills.",
24
  "Neural model weights and MANO assets are not included; precomputed reconstruction is included.",
25
  "Execution receipts retain original paths as provenance; use a fresh output directory for new runs."
26
  ]
27
+ }
agentic_pipeline/2026-10-06/reports/hand_only_sharpa_verification.json DELETED
@@ -1,115 +0,0 @@
1
- {
2
- "dataset": "EgoVerse",
3
- "mode": "hand_only",
4
- "task": "fold_clothes_left",
5
- "robot": "sharpa",
6
- "side": "left",
7
- "status": "complete_hand_only_ik_verified",
8
- "source_frame_range": [
9
- 90,
10
- 210
11
- ],
12
- "frames": 120,
13
- "fps": 30.0,
14
- "interval_duration_seconds": 3.966666666666667,
15
- "source_observations_preserved": true,
16
- "qpos_shape": [
17
- 120,
18
- 35
19
- ],
20
- "qvel_shape": [
21
- 120,
22
- 34
23
- ],
24
- "all_saved_states_finite": true,
25
- "object_geometry_count": 0,
26
- "object_pose_semantics": "identity inertial placeholder, not tracked object",
27
- "frame_convention": "SLAM frame retained; gravity not inferred",
28
- "centering_translation": [
29
- 0.05506010353565216,
30
- 0.24227192997932434,
31
- 1.038370132446289
32
- ],
33
- "fk_position_residuals": {
34
- "wrist": {
35
- "mean_m": 0.026626141797745927,
36
- "max_m": 0.0605165312344021
37
- },
38
- "thumb": {
39
- "mean_m": 0.007315466043140337,
40
- "max_m": 0.0453949752690483
41
- },
42
- "index": {
43
- "mean_m": 0.009320743319975613,
44
- "max_m": 0.03278468643740826
45
- },
46
- "middle": {
47
- "mean_m": 0.007890614852906687,
48
- "max_m": 0.027306433142362956
49
- },
50
- "ring": {
51
- "mean_m": 0.0029973681103229576,
52
- "max_m": 0.017158009906200783
53
- },
54
- "pinky": {
55
- "mean_m": 0.006052517747857886,
56
- "max_m": 0.029134970965542652
57
- }
58
- },
59
- "wrist_orientation_residual_rad": {
60
- "mean": 0.031322445642159885,
61
- "max": 0.06664044237668286
62
- },
63
- "scene_builder": "examples.egoverse.hand_scene",
64
- "joint_limit_overshoot": {},
65
- "robot_fingers_without_source_target": [],
66
- "unretargeted_human_fingers": [],
67
- "cloth_simulation": false,
68
- "object_manipulation_simulation": false,
69
- "manipulation_success": null,
70
- "ik_rollout_note": "Shared IK also writes a noisy forward-rollout diagnostic; it is not a scored manipulation trajectory.",
71
- "files": [
72
- {
73
- "path": "processed/egoverse_hand/sharpa/left/fold_clothes_left/scene.xml",
74
- "sha256": "e2c1b74d188ec3e2c1a133457e9aa067155ad7fa12deef77c6ee9b4069d8f7b5"
75
- },
76
- {
77
- "path": "processed/egoverse_hand/sharpa/left/fold_clothes_left/0/trajectory_kinematic.npz",
78
- "sha256": "fb8aafe61241e8efc9866e9c82ecb0306394c0cc497f3a0b6ffbb3d7b3c9c20a"
79
- },
80
- {
81
- "path": "processed/egoverse_hand/mano/left/fold_clothes_left/0/source_observations.npz",
82
- "sha256": "5cd830281514874d4f2c6f142108c286cab8dc83121a323028cbbf9e20399f05"
83
- },
84
- {
85
- "path": "processed/egoverse_hand/mano/left/fold_clothes_left/0/trajectory_keypoints.npz",
86
- "sha256": "d64d759ba2107cb562d761613ea39026115922c14c3f1997d516d3793d03d21b"
87
- },
88
- {
89
- "path": "processed/egoverse_hand/mano/left/fold_clothes_left/egoverse_provenance.json",
90
- "sha256": "375a6d292dd0709fe241da1d9166a82a76a2ee0430a48e9072bccd5c6fa7b9a3"
91
- },
92
- {
93
- "path": "processed/egoverse_hand/sharpa/left/fold_clothes_left/task_info.json",
94
- "sha256": "c63a6e6d5edba47ca7ce397b993f7d75fcf916a3661aea471959c71f685cc3c1"
95
- },
96
- {
97
- "path": "processed/egoverse_hand/sharpa/left/fold_clothes_left/scene_provenance.json",
98
- "sha256": "e1e61c99003ad785d8dfdd9ced1d5747aa81170f883c2a1f8d637e428cb59dce"
99
- }
100
- ],
101
- "historical_baseline_preservation": {
102
- "verified_files": 17,
103
- "changed_files": [],
104
- "baseline_robot": "allegro"
105
- },
106
- "final_local_checks": {
107
- "test_command": "uv run --no-sync --with 'zarr>=3,<4' -m pytest -q spider/process_datasets/egoverse_test.py",
108
- "passed": 15,
109
- "failed": 0,
110
- "ruff_check": "passed",
111
- "ruff_format_check": "passed",
112
- "current_SPIDER_loader": "passed",
113
- "gpu_execution": false
114
- }
115
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
agentic_pipeline/2026-10-06/reports/object_archive_tests.txt ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ........................................................................ [ 31%]
2
+ ......................................................... [ 55%]
3
+ ........................................................................ [ 86%]
4
+ ............................... [100%]
5
+ =============================== warnings summary ===============================
6
+ ../../../local/home/zhenyuj/spider/.venv/lib/python3.12/site-packages/torch/jit/_script.py:365: 14 warnings
7
+ /local/home/zhenyuj/spider/.venv/lib/python3.12/site-packages/torch/jit/_script.py:365: DeprecationWarning: `torch.jit.script_method` is deprecated. Please switch to `torch.compile` or `torch.export`.
8
+ warnings.warn(
9
+
10
+ -- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
11
+ 232 passed, 14 warnings, 21 subtests passed in 27.87s
agentic_pipeline/2026-10-06/reports/sharpa_gpu/gpu_verification.json CHANGED
@@ -7,14 +7,5 @@
7
  "worlds": 2,
8
  "steps": 10,
9
  "rendered": true
10
- },
11
- {
12
- "dataset": "egoverse_hand",
13
- "task": "fold_clothes_left",
14
- "robot_type": "sharpa",
15
- "status": "passed",
16
- "worlds": 2,
17
- "steps": 10,
18
- "rendered": true
19
  }
20
- ]
 
7
  "worlds": 2,
8
  "steps": 10,
9
  "rendered": true
 
 
 
 
 
 
 
 
 
10
  }
11
+ ]
agentic_pipeline/2026-10-06/reports/sharpa_gpu_verification.json DELETED
@@ -1,34 +0,0 @@
1
- {
2
- "robot": "sharpa",
3
- "dataset": "egoverse_hand",
4
- "task": "fold_clothes_left",
5
- "status": "passed",
6
- "scope": "four-step hand-only optimization integration check",
7
- "metrics": {
8
- "tracking_domain": "hand",
9
- "saved_frames": 4,
10
- "completed_sim_steps": 4,
11
- "completion_fraction": 1.0,
12
- "source_completion_fraction": 0.010084033613445379,
13
- "tracking_pos_err": 0.002843451283395799,
14
- "tracking_rot_err": 0.003723327554406844,
15
- "hand_pos_err": 0.002843451283395799,
16
- "hand_rot_err": 0.003723327554406844,
17
- "max_pos_error_m": 0.004698051005028858,
18
- "max_rot_error_rad": 0.006555105726881763,
19
- "tracking_within_thresholds": true,
20
- "position_threshold_m": 0.02,
21
- "rotation_threshold_rad": 0.2,
22
- "per_hand": {
23
- "left": {
24
- "mean_site_error_m": 0.002843451283395799,
25
- "max_site_error_m": 0.004698051005028858,
26
- "mean_wrist_orientation_error_rad": 0.003723327554406844
27
- }
28
- },
29
- "per_object": {},
30
- "manipulation_success": null,
31
- "definition": "Robot palm/fingertip FK position and wrist orientation versus the IK reference; no object/manipulation score."
32
- },
33
- "manipulation_success": null
34
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
agentic_pipeline/2026-10-06/source.tar.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3d657049c26d73de278fb4e037edd7a7f42b2721a2a5262c523b83d1f6718826
3
- size 183064835
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2724c4dad8b172fec0bbc11f54a636345c4289ca397574d5fec99bf88e91525c
3
+ size 183276422