File size: 103,683 Bytes
578663d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 | [
{
"catalog_number": 1,
"scene_key": "allen_bolt",
"official_scene_id": "assembly.allen_bolt",
"category": "Assembly",
"name": "Allen bolt",
"instruction": "Drive an Allen bolt into a threaded plate using an L-shaped key.",
"difficulty": "Medium",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "key_grasped",
"weight": 1,
"normalized_weight": 0.5,
"mode": "once"
},
{
"stage": "driven",
"weight": 1,
"normalized_weight": 0.5,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the allen_bolt scene: every bolt driven home with the key.\n\nRubric stages (weights and modes live in RUBRIC):\n key_grasped a key has been taken in the weld-on-closure grip (`scene.grasp_held`, any\n key, either arm) — the scene's own grasp state; a milestone, latched. Also\n true once every bolt is seated (a seated bolt was necessarily driven by\n the key), so the rung nests under `driven`. Under an embodiment with no\n gripper (the weld contract disabled, e.g. robot=\"null\") the grasp state does\n not exist and the rung reads from that implication alone\n driven mean over bolts of the tip's depth from the start register to the seat —\n `bolt_stage_depth` -> `seat_depth` when the bolts spawn hand-started\n (`bolt_staged`, every robot binding), plate top (depth 0) -> `seat_depth`\n for the lying spawn — counted only while the bolt is within `align_xy` of the\n hole axis and `align_axis_deg` of upright, the scene's own non-depth seat\n gates, so a bolt's value is 1.0 iff the scene calls it seated\nBoth stages are per env. The key's own \"seated in the hex socket\" step is NOT a rung: the\nhead/socket geometry is not in the scene cfg, so it cannot be read from state without an\ninvented tolerance. Success is the scene's `success()` per env (every bolt `seated()`),\nread when the delivery finishes.",
"grader_summary": "Every bolt seated — driven down to seat depth with the key, upright on the hole axis.\n\nLadder: key_grasped 0.50 · driven 1.00 — two equal rungs for the two measurable steps\n(take the key, drive the bolt home); the second is a linear ramp of tip depth from the\nbolt's start register (`bolt_stage_depth`, or the plate top for a lying spawn) to\n`seat_depth`. With N bolts `driven` is the mean over bolts.",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/assembly/grader/allen_bolt_assembly.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/assembly/grader/allen_bolt_assembly.py",
"gallery": {
"path": "media/allen_bolt/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/allen_bolt.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "842d7c59cfc60edd4db132d1483627096a03ed9376c93d1b59652bdebaeec134",
"bytes": 384758
},
"dataset_source": null
},
{
"catalog_number": 2,
"scene_key": "bulb",
"official_scene_id": "assembly.bulb",
"category": "Assembly",
"name": "Bulb screwing",
"instruction": "Pick up a light bulb and screw it into its socket until fully seated.",
"difficulty": "Medium",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "lifted",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "once"
},
{
"stage": "engaged",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
},
{
"stage": "threaded",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the bulb scene: every bulb seated.\n\nRubric stages (weights and modes live in RUBRIC), each a fraction over that env's bulbs:\n lifted bulbs ever raised to the socket's bore-mouth height — origin at/above\n `socket_opening_z` in the nearest socket's frame, the height a cap-down bulb\n has to clear before it can be set on the socket — or already in a bore (a bulb\n in the bore was necessarily lifted there); a milestone, latched per bulb\n engaged bulbs on/in a socket bore — within `align_xy` of the axis and at/below the\n bore mouth (`socket_opening_z`)\n threaded mean thread depth over the scene's own glow band, from `light_start_z` (free\n rest on the thread, where the glow begins) to `seat_z` (seated), counted only\n while the bulb is aligned in xy AND tilt exactly as the scene's `seated()`\n demands — so a bulb's value is 1.0 iff the scene calls it seated\nThe rungs nest (seated => engaged => lifted), so progress reaches 1.0 exactly when every\nbulb of the env is seated, and for a grader built after the pick as well as one that\nwatched it. Success is the scene's `seated()` per env, read when the delivery finishes.",
"grader_summary": "Every bulb seated — threaded to depth, on the socket axis, upright.\n\nLadder: lifted 0.33 · engaged 0.67 · threaded 1.00 — three equal rungs for the three\nsteps of the task (pick the bulb up, set it on the socket, screw it home); the last\nrung is a linear ramp of thread depth over the scene's own glow band\n(`light_start_z` -> `seat_z`). With N bulbs every stage is the fraction of bulbs.",
"success_source": "def check_success(self):\n return self.scene.seated().all(dim=1)",
"grader_source_path": "robobench/suites/assembly/grader/bulb_assembly.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/assembly/grader/bulb_assembly.py",
"gallery": {
"path": "media/bulb/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/bulb.mp4",
"website_playback_rate": 5.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "0643ff21afba057dff61af5e35411ddaf9552e794a730cad9ed953d83de4f6eb",
"bytes": 295291
},
"dataset_source": {
"repo_id": "EmbodiedSWE/bulb_franka_pinkik.jointtarget_15hz_generalization_train_20260921",
"revision": "53035d12cccc88f1d68a8fb75a12092e8739952e",
"format": "v3.0",
"fps": 15,
"total_episodes": 816,
"camera_order": [
"front",
"wrist"
],
"task_index_scope": "local to this source repository; not a benchmark-global task ID",
"robot_type": "assembly.bulb.franka.pink_ik",
"selected_episodes": [
0,
408,
815
],
"source_license_metadata": "None"
}
},
{
"catalog_number": 3,
"scene_key": "ikea_table",
"official_scene_id": "assembly.ikea_table",
"category": "Assembly",
"name": "Table assembly",
"instruction": "Attach four threaded legs to the corner studs of a tabletop.",
"difficulty": "Hard",
"catalog_embodiments": "Bimanual Franka",
"rubric": [
{
"stage": "aligned",
"weight": 1,
"normalized_weight": 0.5,
"mode": "live"
},
{
"stage": "seated",
"weight": 1,
"normalized_weight": 0.5,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the ikea_table scene: all four legs screwed down onto the tabletop's studs.\n\nThe four legs are identical sub-goals, so every rung is ONE stage whose value is the fraction\nof legs (k/4) that have reached it. The scene has no grasp contract and exposes no threshold\nfor \"lifted\", so a grasped/lifted rung cannot be measured from state and is not one; per leg\nthe ladder is aligned -> seated, equal weights, all measured in the tabletop's own frame\n(order-independent: any leg on any stud, as the scene's `seated()` is):\n aligned legs standing on a stud — within `align_xy` of the nearest stud and screw axis\n within `align_axis_deg` of the stud axis: the scene's own two non-depth seat gates\n seated mean over legs of the screw-down from the stud-top rest height (`leg_start_z`, the\n scene's on-stud spawn height) to the seat (`seat_z`), counted only while aligned —\n so a leg's value is 1.0 iff the scene calls it `seated()`\nThe rungs nest (seated => aligned), so progress reaches 1.0 exactly when every leg is seated.\nThe scene welds a seated leg to the table, so a welded leg stays `seated()` however the\nassembly is later moved. Success is the scene's own `success()` (every leg `seated()`), read\nwhen the delivery finishes.",
"grader_summary": "All four legs seated on the studs — each threaded down to depth, on a stud axis, upright.\n\nLadder (per leg, each rung 1/8): aligned 1/8 · seated 2/8;\n1 leg 0.25 · 2 legs 0.50 · 3 legs 0.75 · 4 legs 1.00 — two equal rungs for the two\nmeasurable steps of each leg (set it on its stud, screw it down); the second is a linear\nramp of the leg's height from the stud-top rest height (`leg_start_z`) to the seat (`seat_z`).",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/assembly/grader/ikea_table_assembly.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/assembly/grader/ikea_table_assembly.py",
"gallery": {
"path": "media/ikea_table/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/ikea_table.mp4",
"website_playback_rate": 4.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "5f68332685fb6baacb949ab2bac84a89066ffee12682f63d1143613630a100c5",
"bytes": 1151335
},
"dataset_source": null
},
{
"catalog_number": 4,
"scene_key": "so101",
"official_scene_id": "assembly.so101",
"category": "Assembly",
"name": "Robot-arm assembly",
"instruction": "Assemble a robot arm.",
"difficulty": "Hard",
"catalog_embodiments": "Bimanual Franka",
"rubric": [
{
"stage": "motor_seated",
"weight": 1,
"normalized_weight": 0.07142857142857142,
"mode": "live"
},
{
"stage": "tab_screws_picked",
"weight": 2,
"normalized_weight": 0.14285714285714285,
"mode": "once"
},
{
"stage": "tab_screws_fastened",
"weight": 2,
"normalized_weight": 0.14285714285714285,
"mode": "live"
},
{
"stage": "fork_seated",
"weight": 1,
"normalized_weight": 0.07142857142857142,
"mode": "live"
},
{
"stage": "horn_screws_picked",
"weight": 4,
"normalized_weight": 0.2857142857142857,
"mode": "once"
},
{
"stage": "horn_screws_fastened",
"weight": 4,
"normalized_weight": 0.2857142857142857,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the so101 scene: the elbow servo seated + tab-screwed into the upper arm, the forearm\nclipped onto its horn + horn-screwed.\n\nThe scene defines no `success()`; the criterion below is read off its own assembly geometry\nand fastening state (see `check_success`):\n every loose screw fastened — the four identical M2 tab screws in the four tab holes and the\n eight identical M3 horn screws in the eight horn-line holes (the scene's own `state == 2`,\n \"fully fastened\") — with the servo live in the arm pocket and the forearm live on the horn.\n\nRubric stages (weights and modes live in RUBRIC):\n motor_seated the bare servo sits in the upper arm's pocket: its body frame on the\n upper_arm link frame (identical by construction when seated) within the\n scene's fastening-gate tolerances `motor_align_pos` / `motor_align_deg`\n — the scene's own PARTS ALIGNED rule for the tab holes\n tab_screws_picked fraction of the four identical M2 tab screws ever carried on the drill's\n magnetic bit (`attached`, the scene's only \"grasp\" of a screw; a\n fastened screw was necessarily driven off that bit) — a milestone\n tab_screws_fastened fraction of the four M2s welded home in a tab hole (`fastened`)\n fork_seated the forearm's clevis sits on the servo's output horn: the lower_arm at\n the elbow-joint seat transform of the motor, within the same\n `motor_align_pos` / `motor_align_deg` gate, angle-agnostic about the\n horn axis (the assembled elbow may stand at any angle) — the scene's\n PARTS ALIGNED rule for the horn holes\n horn_screws_picked fraction of the eight identical M3 horn screws ever carried on the bit\n horn_screws_fastened fraction of the eight M3s welded home in a horn-line hole\nEvery stage is per env. The two \"picked\" milestones are the grasped rung of the screws' ladder;\nthe servo and the forearm have no grasp the scene can measure (no grasp contract, no lift\nthreshold in cfg), so their ladders start at the seated rung. A screw's transient \"in the hole,\nnot yet home\" state is not a rung: the auto-driver takes it from engaged to fastened by itself.",
"grader_summary": "The SO101 elbow assembled: servo seated and tab-screwed into the upper arm, forearm clipped onto the horn and horn-screwed.\n\nLadder (14 units — one unit per part: servo 1, 4 M2 tab screws 4, forearm 1, 8 M3 horn\nscrews 8; a screw's unit split equally over its two rungs, picked 0.5 + fastened 0.5):\n servo in pocket 1/14 = 0.071 · 4 M2 on the bit 3/14 = 0.214 · 4 M2 fastened 5/14 = 0.357 ·\n forearm on horn 6/14 = 0.429 · 8 M3 on the bit 10/14 = 0.714 · 8 M3 fastened 14/14 = 1.000\nThe identical screws of each group share one k/N fraction stage weighted N/2 per rung.",
"success_source": "def check_success(self):\n \"\"\"Every loose screw fastened (the scene's `state == 2`) AND the servo live in the\n pocket AND the forearm live on the horn — the assembled elbow, on the final state.\"\"\"\n all_fastened = (self.scene.fastened >= 0).all(dim=1)\n return all_fastened & self._motor_seated() & self._fork_seated()",
"grader_source_path": "robobench/suites/assembly/grader/so101_assembly.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/assembly/grader/so101_assembly.py",
"gallery": {
"path": "media/so101/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/so101.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "e11560d36e25e18a217b14098189c598b870978a562bfe6d3c16dae268af3a0f",
"bytes": 928445
},
"dataset_source": null
},
{
"catalog_number": 5,
"scene_key": "nut_thread",
"official_scene_id": "assembly.nut_thread",
"category": "Assembly",
"name": "Nut threading",
"instruction": "Pick up an M16 nut and thread it onto a fixed bolt.",
"difficulty": "Easy",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "lifted",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "once"
},
{
"stage": "aligned",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
},
{
"stage": "threaded",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the nut_thread scene: every nut seated on a bolt.\n\nRubric stages (weights and modes live in RUBRIC), each a fraction over that env's nuts:\n lifted nuts ever raised to the bolt-top rest height — origin at/above `bolt_height`\n in the nearest bolt's frame, the height the scene quotes for a nut resting on\n the bolt top and so the height a nut has to clear before it can be set on the\n bolt — or already aligned on a bolt (a nut on the bolt was necessarily lifted\n there); a milestone, latched per nut\n aligned nuts aligned on a bolt — within `align_xy` of the bolt axis and tilted at most\n `align_axis_deg` off it, the scene's own two non-depth seat gates\n threaded mean thread depth from the bolt-top rest height (`bolt_height`) down to the\n seat (`seat_z`), counted only while aligned — so a nut's value is 1.0 iff the\n scene calls it seated\nThe rungs nest (seated => aligned => lifted), so progress reaches 1.0 exactly when every\nnut of the env is seated, and for a grader built after the pick as well as one that watched\nit. Success is the scene's `seated()` per env, read when the delivery finishes.",
"grader_summary": "Every nut seated — threaded down to seat depth, on the bolt axis, upright.\n\nLadder: lifted 0.33 · aligned 0.67 · threaded 1.00 — three equal rungs for the three\nsteps of the task (pick the nut up, set it on the bolt, screw it down); the last rung\nis a linear ramp of the nut's height from the bolt-top rest height (`bolt_height`) to\nthe seat (`seat_z`). With N nuts every stage is the fraction of nuts.",
"success_source": "def check_success(self):\n return self.scene.seated().all(dim=1)",
"grader_source_path": "robobench/suites/assembly/grader/nut_thread_assembly.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/assembly/grader/nut_thread_assembly.py",
"gallery": {
"path": "media/nut_thread/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/nut_thread.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "23ea1300efdc4a4b63d272143b44b60c10c74a4bcd9e5e7ef190d22b020200d3",
"bytes": 178375
},
"dataset_source": {
"repo_id": "EmbodiedSWE/nut_thread_franka_ik.jointtarget_20hz_20260905",
"revision": "7c9282badf83850cca7c56159f2e048ad62cb93a",
"format": "v3.0",
"fps": 20,
"total_episodes": 401,
"camera_order": [
"front",
"side",
"wrist"
],
"task_index_scope": "local to this source repository; not a benchmark-global task ID",
"robot_type": "assembly.nut_thread.franka.joint",
"selected_episodes": [
0,
200,
400
],
"source_license_metadata": "license: apache-2.0\ntask_categories:\n- robotics\ntags:\n- LeRobot\n- robotics\n- franka\n- assembly\n- nut-threading\n- isaac-sim\n- cosigen\n- sim\nconfigs:\n- config_name: default\n data_files: data/*/*.parquet"
}
},
{
"catalog_number": 6,
"scene_key": "pc_gpu",
"official_scene_id": "assembly.pc_gpu",
"category": "Assembly",
"name": "GPU insertion",
"instruction": "Align a graphics card with a PCIe slot and press it into place.",
"difficulty": "Easy",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "grasped",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "once"
},
{
"stage": "aligned",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
},
{
"stage": "pressed",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the pc_gpu scene: the graphics card seated in the PCIe x16 slot.\n\nRubric stages (weights and modes live in RUBRIC):\n grasped the card has been taken in the weld-on-closure grip (`scene.grasp_held`) — the\n scene's own grasp state; a milestone, latched. Also true once the card is\n aligned in the slot (a card in the slot was necessarily carried there), so the\n rung nests under `aligned`. Under an embodiment with no gripper (the weld\n contract disabled, e.g. robot=\"null\") the grasp state does not exist and the\n rung reads from that implication alone\n aligned the card lined up on the slot — origin within `align_xy` of the seated point,\n up axis within `align_axis_deg` of the slot axis, length axis within\n `align_yaw_deg` of the slot heading: the scene's own three non-depth seat gates\n pressed tab depth below the slot mouth (`slot_mouth_z`, the scene's own `engaged()`)\n as a fraction of `seat_depth`, counted only while aligned — so it is 1.0 iff\n the scene calls the card seated\nEvery stage is per env. The rungs nest (seated => aligned => grasped), so progress reaches\n1.0 exactly when the card is seated, and for a grader built after the pick as well as one\nthat watched it. The rear-panel pass-through is NOT a rung: the cutout's geometry is not in\nthe scene cfg, so it cannot be read from state without an invented tolerance. Success is\nthe scene's `seated()` per env — depth, xy, tilt AND heading gates — read when the\ndelivery finishes.",
"grader_summary": "The card seated in its slot — pressed to depth, centred, upright, heading along the slot.\n\nLadder: grasped 0.33 · aligned 0.67 · pressed 1.00 — three equal rungs for the three\nsteps of the task (take the card, line it up on the slot, press it home); the last rung\nis a linear ramp of tab depth from the slot mouth (`slot_mouth_z`) to `seat_depth`.",
"success_source": "def check_success(self):\n return self.scene.seated()",
"grader_source_path": "robobench/suites/assembly/grader/pc_gpu_assembly.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/assembly/grader/pc_gpu_assembly.py",
"gallery": {
"path": "media/pc_gpu/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/pc_gpu.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "ccb137b362404bc9991eca23a8b1aa857b66bec80a3f5240490ef3f05f95878c",
"bytes": 225359
},
"dataset_source": {
"repo_id": "EmbodiedSWE/pc_gpu_franka_osc.jointpos_15hz_20260907",
"revision": "d9d2cb6eb8288a2716758bb3790fe0d37798990b",
"format": "v3.0",
"fps": 15,
"total_episodes": 600,
"camera_order": [
"front",
"wrist"
],
"task_index_scope": "local to this source repository; not a benchmark-global task ID",
"robot_type": "assembly.pc_gpu.franka.osc",
"selected_episodes": [
0,
49,
114
],
"source_license_metadata": "license: apache-2.0\ntask_categories:\n- robotics\ntags:\n- LeRobot\n- franka\n- assembly\n- pc-gpu\n- isaac-sim\n- cosigen\n- sim\n- data-gen\nconfigs:\n- config_name: default\n data_files: data/*/*.parquet"
}
},
{
"catalog_number": 7,
"scene_key": "pc_gpu_ram",
"official_scene_id": "assembly.pc_gpu_ram",
"category": "Assembly",
"name": "GPU + RAM install",
"instruction": "Install a graphics card and then seat two memory modules in their slots.",
"difficulty": "Medium",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "gpu_grasped",
"weight": 1,
"normalized_weight": 0.1111111111111111,
"mode": "once"
},
{
"stage": "gpu_aligned",
"weight": 1,
"normalized_weight": 0.1111111111111111,
"mode": "live"
},
{
"stage": "gpu_seated",
"weight": 1,
"normalized_weight": 0.1111111111111111,
"mode": "live"
},
{
"stage": "ram_grasped",
"weight": 2,
"normalized_weight": 0.2222222222222222,
"mode": "once"
},
{
"stage": "ram_aligned",
"weight": 2,
"normalized_weight": 0.2222222222222222,
"mode": "live"
},
{
"stage": "ram_seated",
"weight": 2,
"normalized_weight": 0.2222222222222222,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the pc_gpu_ram scene: the graphics card AND both RAM sticks seated.\n\nThree parts — one card, two identical sticks. Each part climbs the same ladder, grasped ->\naligned -> seated, and every part is worth the same third of the score. The card has its own\nthree stages; the two identical sticks share three FRACTION stages (k/2 each), which\ntherefore carry weight 2 (one unit per stick) against the card's 1.\n gpu_grasped / ram_grasped parts ever taken in the weld-on-closure grip (`scene.grasp_held`:\n column 0 the card, 1.. the sticks, in `grasp_sites()` order) —\n the scene's own grasp state; milestones, latched per part. Also\n true for a part that is seated (a seated part was necessarily\n carried there), so the rungs nest under the seated rungs. Under\n an embodiment with no gripper (the weld contract disabled, e.g.\n robot=\"null\") the grasp state does not exist and the rungs read\n from that implication alone\n gpu_aligned / ram_aligned the part lined up on its slot — origin within `gpu_align_xy` /\n `ram_align_xy` of the seated point, up axis within\n `gpu_align_axis_deg` / `ram_align_axis_deg` of the slot axis,\n length axis within `gpu_align_yaw_deg` / `ram_align_yaw_deg` of\n the slot heading: the scene's own three non-depth seat gates\n gpu_seated / ram_seated depth below the slot mouth (`gpu_slot_mouth_z` / `ram_slot_mouth_z`,\n the scene's own `gpu_engaged()` / `ram_engaged()`) as a fraction\n of `gpu_seat_depth` / `ram_seat_depth`, counted only while\n aligned — so a part's value is 1.0 iff the scene calls it seated\nThe rungs nest (seated => aligned => grasped), so progress reaches 1.0 exactly when the card\nand both sticks are seated, and for a grader built after the picks as well as one that\nwatched them. Success is the scene's own `success()` (card and every stick seated), read when\nthe delivery finishes.",
"grader_summary": "The graphics card and both RAM sticks seated — each pressed to depth, centred, upright, heading along its slot.\n\nLadder (total weight 9, every part worth 3/9):\ncard grasped 1/9 · card aligned 2/9 · card seated 3/9 (0.333) ·\nthen per stick +1/9 per rung: first stick seated 6/9 (0.667) · both sticks seated 1.00.\nEach part's seated rung is a linear ramp of depth from its slot mouth to its seat depth.",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/assembly/grader/pc_gpu_ram_assembly.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/assembly/grader/pc_gpu_ram_assembly.py",
"gallery": {
"path": "media/pc_gpu_ram/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/pc_gpu_ram.mp4",
"website_playback_rate": 2.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "014f476acf2f933a70997d8bcead427d47ce886d0cf1569f0a7b8be331f5d31f",
"bytes": 809933
},
"dataset_source": {
"repo_id": "EmbodiedSWE/pc_gpu_ram_franka_osc.jointpos_15hz_20260907",
"revision": "29f0af1fa2c33b3572f6b80c44a7eb4ed2dd4585",
"format": "v3.0",
"fps": 15,
"total_episodes": 600,
"camera_order": [
"front",
"high",
"wrist"
],
"task_index_scope": "local to this source repository; not a benchmark-global task ID",
"robot_type": "assembly.pc_gpu_ram.franka.osc",
"selected_episodes": [
0,
16,
84
],
"source_license_metadata": "license: apache-2.0\ntask_categories:\n- robotics\ntags:\n- LeRobot\n- franka\n- assembly\n- pc-gpu-ram\n- isaac-sim\n- cosigen\n- sim\n- data-gen\nconfigs:\n- config_name: default\n data_files: data/*/*.parquet"
}
},
{
"catalog_number": 8,
"scene_key": "pc_motherboard",
"official_scene_id": "assembly.pc_motherboard",
"category": "Assembly",
"name": "Motherboard mount",
"instruction": "Use an Allen key to drive seven mounting bolts and secure a motherboard inside a case.",
"difficulty": "Medium",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "key_grasped",
"weight": 1,
"normalized_weight": 0.125,
"mode": "once"
},
{
"stage": "driven",
"weight": 7,
"normalized_weight": 0.875,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the pc_motherboard scene: all seven case-mount bolts driven home with the key.\n\nRubric stages (weights and modes live in RUBRIC):\n key_grasped the key has been taken in the weld-on-closure grip (`scene.grasp_held`, whose\n only site is the key handle) — the scene's own grasp state; a milestone,\n latched. Also true once every bolt is seated (a seated bolt was necessarily\n driven by the key), so the rung nests under `driven`. Under an embodiment\n with no gripper (the weld contract disabled, e.g. robot=\"null\") the grasp\n state does not exist and the rung reads from that implication alone\n driven mean over the seven bolts of the tip's depth from the start register to the\n seat — `stage_depth` -> `seat_depth` under the screw mechanic (every bolt\n spawns hand-started that deep), board face (depth 0) -> `seat_depth` when\n the mechanic is off and a loose bolt is threaded in from the top — counted\n only while the bolt is within `align_xy` of its nearest hole axis and\n `align_axis_deg` of upright, the scene's own non-depth seat gates, so a\n bolt's value is 1.0 iff the scene calls it `seated()`\nBoth stages are per env. The bolts spawn pre-staged in their holes, so an \"aligned\" rung\nwould be free at reset and is not one; the key's own \"seated in the hex socket\" step is NOT a\nrung either: the head/socket geometry is not in the scene cfg, so it cannot be read from\nstate without an invented tolerance. Success is the scene's own `success()` (every bolt\n`seated()`), read when the delivery finishes.",
"grader_summary": "All seven motherboard bolts seated — each driven to depth on its hole axis, upright.\n\nLadder (total weight 8, one unit per part — the key and each of the 7 bolts):\nkey grasped 0.125 · 1 bolt 0.25 · 2 bolts 0.375 · 3 bolts 0.50 · 4 bolts 0.625 ·\n5 bolts 0.75 · 6 bolts 0.875 · 7 bolts 1.00. `driven` is the mean over bolts of a linear\nramp of tip depth from the start register (`stage_depth`, or the board face for a loose\nspawn) to `seat_depth`.",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/assembly/grader/pc_motherboard_assembly.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/assembly/grader/pc_motherboard_assembly.py",
"gallery": {
"path": "media/pc_motherboard/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/pc_motherboard.mp4",
"website_playback_rate": 2.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "60316e66d9158090fd81dbcbb682e4fb3043dc9899f9f55803eae885654c5b73",
"bytes": 379718
},
"dataset_source": null
},
{
"catalog_number": 9,
"scene_key": "pc_ram",
"official_scene_id": "assembly.pc_ram",
"category": "Assembly",
"name": "RAM seating",
"instruction": "Align and seat two memory modules in separate DIMM slots.",
"difficulty": "Easy",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "grasped",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "once"
},
{
"stage": "aligned",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
},
{
"stage": "seated",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the pc_ram scene: both RAM sticks seated in their DIMM slots.\n\nThe two sticks are identical sub-goals, so every rung is ONE stage whose value is the\nfraction of sticks (k/2) that have reached it. Per stick the ladder is grasped -> aligned ->\nseated, equal weights:\n grasped sticks ever taken in the weld-on-closure grip (`scene.grasp_held`, one column\n per stick) — the scene's own grasp state; a milestone, latched per stick. Also\n true for a stick that is seated (a seated stick was necessarily carried there),\n so the rung nests under `seated`. Under an embodiment with no gripper (the weld\n contract disabled, e.g. robot=\"null\") the grasp state does not exist and the\n rung reads from that implication alone\n aligned sticks lined up on their slot — origin within `align_xy` of the seated point, up\n axis within `align_axis_deg` of the slot axis, length axis within `align_yaw_deg`\n of the slot heading: the scene's own three non-depth seat gates\n seated mean over sticks of the blade depth below the slot mouth (`slot_mouth_z`, the\n scene's own `engaged()`) as a fraction of `seat_depth`, counted only while\n aligned — so a stick's value is 1.0 iff the scene calls it `seated()`\nThe rungs nest (seated => aligned => grasped), so progress reaches 1.0 exactly when both\nsticks are seated, and for a grader built after the picks as well as one that watched them.\nSuccess is the scene's own `success()` (every stick `seated()`), read when the delivery\nfinishes.",
"grader_summary": "Both sticks seated — pressed to depth, centred, upright, heading along the slot.\n\nLadder (per stick, each rung 1/6): grasped 1/6 · aligned 2/6 · seated 3/6;\none stick done 0.50 · both sticks done 1.00 — three equal rungs for the three steps of\neach stick (take it, line it up on its slot, press it home); the last rung is a linear\nramp of blade depth from the slot mouth (`slot_mouth_z`) to `seat_depth`.",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/assembly/grader/pc_ram_assembly.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/assembly/grader/pc_ram_assembly.py",
"gallery": {
"path": "media/pc_ram/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/pc_ram.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "161ab877358848f66ee4900b4bc1bf3f9dd3589319164bfdaac755ea4b0eaeea",
"bytes": 261521
},
"dataset_source": {
"repo_id": "EmbodiedSWE/pc_ram_franka_osc.jointpos_15hz_20260907",
"revision": "d2cd033771eb621a2f35e7e9b9a1c7ab943f5a69",
"format": "v3.0",
"fps": 15,
"total_episodes": 600,
"camera_order": [
"front",
"wrist"
],
"task_index_scope": "local to this source repository; not a benchmark-global task ID",
"robot_type": "assembly.pc_ram.franka.osc",
"selected_episodes": [
0,
300,
599
],
"source_license_metadata": "license: apache-2.0\ntask_categories:\n- robotics\ntags:\n- LeRobot\n- franka\n- assembly\n- pc-ram\n- isaac-sim\n- cosigen\n- sim\n- data-gen\nconfigs:\n- config_name: default\n data_files: data/*/*.parquet"
}
},
{
"catalog_number": 10,
"scene_key": "pen_holder",
"official_scene_id": "packing.pen_holder",
"category": "Organizing",
"name": "Pen holder",
"instruction": "Insert every present pen tip-up and then place the filled holder upright.",
"difficulty": "Easy",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "pens_in",
"weight": 0.9,
"normalized_weight": 0.9,
"mode": "live"
},
{
"stage": "holder_placed",
"weight": 0.1,
"normalized_weight": 0.1,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the pen_holder scene: every present pen in the holder tip-up, holder set down upright.\n\nRubric stages (weights and modes live in RUBRIC):\n pens_in fraction of the PRESENT pens the scene counts as in the holder — the\n scene's own `counted()`: bottom within `xy_tol` of the holder axis,\n deeper than `depth_min` below the rim, tip-up within `pen_align_max_deg`,\n holder within `holder_tilt_max_deg` of up, pen and holder settled. All\n pens are one identical family, so they share this one fraction stage;\n the fraction is over the sampled subset (the present pen count is drawn\n per env at reset, k ~ U{min_present..n} per family)\n holder_placed the loaded holder standing upright on the work surface (the scene's\n `holder_placed()`: within `placed_tilt_deg`, bottom within `placed_z_tol`\n of the surface, still), gated on every present pen being in. The holder\n spawns upright on the surface, so ungated it would be free credit at\n reset; the order is fill, then set down\nBoth stages are live: a pen ejected or a holder knocked over loses its credit. Success is\nthe scene's own `success()` — `all_inserted() & holder_placed()` — on the final state.\n\nNot measured: a grasped/lifted rung for the pens. The scene exposes no lift height or hold\ncriterion, and inventing one is forbidden, so that rung is dropped.",
"grader_summary": "Every present pen in the holder tip-up, and the loaded holder standing upright on the surface.\n\nLadder (N present pens): k pens in 0.90 * k / N · holder set down 1.00\n N = 4: 1 pen 0.225 · 2 pens 0.45 · 3 pens 0.675 · 4 pens 0.90 · set down 1.00\n N = 3: 1 pen 0.30 · 2 pens 0.60 · 3 pens 0.90 · set down 1.00",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/packing/grader/pen_holder.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/packing/grader/pen_holder.py",
"gallery": {
"path": "media/pen_holder/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/pen_holder.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "3fb641150508178fbd0900f38122a45364eec92c790e027b142a6254a69466a8",
"bytes": 122980
},
"dataset_source": {
"repo_id": "EmbodiedSWE/pen_holder_franka_diffik.jointtarget_15hz_20260909",
"revision": "c3ecaa765252aaca473cccf7c4e770acc5db7f0e",
"format": "v3.0",
"fps": 15,
"total_episodes": 684,
"camera_order": [
"front",
"corner",
"wrist"
],
"task_index_scope": "local to this source repository; not a benchmark-global task ID",
"robot_type": "packing.pen_holder.franka.diff_ik",
"selected_episodes": [
0,
342,
683
],
"source_license_metadata": "None"
}
},
{
"catalog_number": 11,
"scene_key": "tool_packing",
"official_scene_id": "packing.tool_packing",
"category": "Organizing",
"name": "Tool packing",
"instruction": "Stow three tools in their assigned drawers and close the cabinet.",
"difficulty": "Medium",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "stowed",
"weight": 0.45,
"normalized_weight": 0.45,
"mode": "live"
},
{
"stage": "drawer_shut",
"weight": 0.45,
"normalized_weight": 0.45,
"mode": "live"
},
{
"stage": "cabinet_shut",
"weight": 0.1,
"normalized_weight": 0.1,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the tool_packing scene: three tools stowed in their assigned drawers, cabinet shut.\n\nRubric stages (weights and modes live in RUBRIC):\n stowed fraction of the three tools inside their ASSIGNED drawer's tray — the\n scene's own `stowed()`: origin inside the tray-interior box in the drawer's\n body frame (wherever the drawer slid to), settled\n drawer_shut fraction of the three tools that are stowed AND whose own drawer is shut\n (joint within `drawer_closed_tol` of 0) — the second rung of each tool's\n ladder, since every tool has its own drawer to push home\n cabinet_shut the scene's own `success()`: all three stowed, EVERY drawer shut (the\n distractor drawer included), both doors shut on the toolbox variant, and\n items + joints settled\nAll three stages are live: a tool knocked out of its tray or a drawer that slides back open\nloses its credit. Success is the scene's own `success()` on the final state.\n\nNot measured: a grasped/lifted rung for the tools. The scene exposes no lift height or\nhold criterion, and inventing one is forbidden, so that rung is dropped.",
"grader_summary": "All three tools stowed in their assigned drawers and the cabinet fully shut.\n\nLadder: 1 stowed 0.15 · 2 stowed 0.30 · 3 stowed 0.45 · 1 drawer shut 0.60 ·\n 2 drawers shut 0.75 · 3 drawers shut 0.90 · cabinet shut 1.00",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/packing/grader/tool_packing.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/packing/grader/tool_packing.py",
"gallery": {
"path": "media/tool_packing/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/tool_packing.mp4",
"website_playback_rate": 5.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "b7d769a5d1fd26a206730288507ae77efa806df41a8a630b150f7a60f7214bb2",
"bytes": 708819
},
"dataset_source": {
"repo_id": "EmbodiedSWE/tool_packing_franka_diffik.jointtarget_15hz_20260910",
"revision": "b197c503474a91a911e6b8027f8e50cdfe5d135c",
"format": "v3.0",
"fps": 15,
"total_episodes": 1686,
"camera_order": [
"front",
"corner",
"wrist"
],
"task_index_scope": "local to this source repository; not a benchmark-global task ID",
"robot_type": "packing.tool_packing.franka.diff_ik",
"selected_episodes": [
0,
843,
1685
],
"source_license_metadata": "None"
}
},
{
"catalog_number": 12,
"scene_key": "egg_carton",
"official_scene_id": "packing.egg_carton",
"category": "Organizing",
"name": "Egg carton",
"instruction": "Seat three eggs upright in the cells of an egg carton, then push the hinged lid closed.",
"difficulty": "Hard",
"catalog_embodiments": "Unitree G1",
"rubric": [
{
"stage": "eggs_seated",
"weight": 0.9,
"normalized_weight": 0.9,
"mode": "live"
},
{
"stage": "lid_closed",
"weight": 0.1,
"normalized_weight": 0.1,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the egg_carton scene: `target_eggs` distinct pockets filled with upright eggs, lid closed.\n\nRubric stages (weights and modes live in RUBRIC):\n eggs_seated fraction of the `target_eggs` pockets occupied by a seated egg — the\n scene's own `occupied()`: egg centre within `seat_xy_tol` of a cavity\n centre in the carton frame, in the `seat_z_min..seat_z_max` band, long axis\n within `egg_tilt_max_deg` of the pocket axis, still. Distinct pockets, so a\n second egg dropped into an occupied pocket earns nothing; the eggs are\n identical, so they share this one fraction stage\n lid_closed the scene's own `success()`: every target pocket filled, the lid within\n `lid_closed_deg` of shut (when `require_lid_closed`), eggs and lid settled.\n Gated on the pockets being filled: shutting the lid on an empty carton is not\n progress (it has to be reopened), and the scene's own rubric only awards the\n lid after the eggs\nBoth stages are live: an egg that topples or a lid that swings back open loses its credit.\nSuccess is the scene's own `success()` on the final state.\n\nNot measured: a grasped/lifted rung for the eggs. The scene exposes no lift height or hold\ncriterion, and inventing one is forbidden, so that rung is dropped.",
"grader_summary": "Three eggs seated upright in three distinct carton pockets, then the lid pushed closed.\n\nLadder: 1 egg 0.30 · 2 eggs 0.60 · 3 eggs 0.90 · lid closed 1.00\n(0.90 * k / target_eggs per k pockets filled; the default target is three.)",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/packing/grader/egg_carton.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/packing/grader/egg_carton.py",
"gallery": {
"path": "media/egg_carton/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/egg_carton.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "9243214c2a7895e3fe55fb0aebf0daea1c0069513c03134ccdc5b32f5d376e1d",
"bytes": 1475271
},
"dataset_source": null
},
{
"catalog_number": 13,
"scene_key": "clear_organic_objects",
"official_scene_id": "packing.clear_organic_objects",
"category": "Organizing",
"name": "Clear organics",
"instruction": "Clear the organic objects on a cluttered table into a bin while leaving the other items in place.",
"difficulty": "Medium",
"catalog_embodiments": "Unitree G1",
"rubric": [
{
"stage": "organics_cleared",
"weight": 1.0,
"normalized_weight": 1.0,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the clear_organic_objects scene: every organic in the bin, nothing else in it.\n\nRubric stages (weights and modes live in RUBRIC):\n organics_cleared fraction of the required placements that hold right now — in the\n legacy variant every PRESENT organic cleared into the bin (the scene's\n `cleared()`: origin inside the bin's interior box in the bin's body\n frame, settled); in the sort variant (`sort_to_bin` / `sort_to_tray`)\n the scene's own `bin_targets` cleared plus its `tray_targets` resting\n on the tray. The organics are interchangeable sub-goals, so they share\n this one fraction stage. It is GATED by the two sanity predicates\n below: the whole stage reads 0 while either fails, and recovers when\n the state is repaired (live)\nGates (no weight of their own — rule 7 sanity predicates on the one and only stage, which\nis also the done stage):\n no distractor in the bin the scene's `distractors_in_bin()` — identification IS the\n task: dumping everything in must not score as progress\n bin upright the bin's up-axis within its own topple angle of world-up, the\n angle (from `bin_bbox` x `bin_scale`) past which a free crate's\n centre leaves its base and it falls onto its side. A knocked-\n over bin with produce inside it must not pass. The default bin\n is kinematic and cannot move, so this only bites in the\n dynamic-bin variant\nSuccess is the scene's own `success()` (every required placement holds, no distractor in the\nbin) AND the bin upright, on the final state. `progress == 1` exactly when that holds.\n\nNot measured: a grasped/lifted rung per item (the scene exposes no lift height or hold\ncriterion) and \"bin in place\" (a positional test would need an invented tolerance; the\nbody-frame judging already makes a moved bin judge identically).",
"grader_summary": "Every organic item in the bin, no non-food item in it, the bin still upright.\n\nLadder (N required placements): k placed k / N — e.g. the default manifest's 15 organics:\n1 in 0.067 · 2 in 0.133 · ... · 14 in 0.933 · 15 in 1.00 (a preset with 11 present: 0.091\nper organic); any non-food item in the bin, or the bin tipped over, gates the stage to 0\nuntil repaired.",
"success_source": "def check_success(self):\n return self.scene.success() & self._bin_upright()",
"grader_source_path": "robobench/suites/packing/grader/clear_organic_objects.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/packing/grader/clear_organic_objects.py",
"gallery": {
"path": "media/clear_organic_objects/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/clear_organic.mp4",
"website_playback_rate": 2.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "f2b90694655ae65a2cffb709cefb4f7288ee2adfcd41fd415a02fe7478ea1000",
"bytes": 694445
},
"dataset_source": {
"repo_id": "EmbodiedSWE/clear_organic_agent_ab_A400.jointtarget_20hz_20260918",
"revision": "c36f831161a17a6515740cd5c74739e1e79e2e3a",
"format": "v3.0",
"fps": 20,
"total_episodes": 400,
"camera_order": [
"front",
"side",
"wrist"
],
"task_index_scope": "local to this source repository; not a benchmark-global task ID",
"robot_type": "packing.clear_organic_objects.franka.joint",
"selected_episodes": [
0,
200,
399
],
"source_license_metadata": "license: apache-2.0\ntask_categories:\n- robotics\ntags:\n- LeRobot\n- robotics\n- franka\n- packing\n- produce\n- pick-and-place\n- isaac-sim\n- cosigen\n- sim\n- agent-ab\nconfigs:\n- config_name: default\n data_files: data/*/*.parquet"
}
},
{
"catalog_number": 14,
"scene_key": "coffee",
"official_scene_id": "puzzle.coffee",
"category": "Puzzle",
"name": "Coffee brewing",
"instruction": "Load a capsule, brew coffee, and return the filled mug to the serving tray.",
"difficulty": "Medium",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "pod_loaded",
"weight": 1,
"normalized_weight": 0.14285714285714285,
"mode": "once"
},
{
"stage": "bay_closed",
"weight": 1,
"normalized_weight": 0.14285714285714285,
"mode": "once"
},
{
"stage": "cup_staged",
"weight": 1,
"normalized_weight": 0.14285714285714285,
"mode": "once"
},
{
"stage": "started",
"weight": 1,
"normalized_weight": 0.14285714285714285,
"mode": "once"
},
{
"stage": "brewed",
"weight": 1,
"normalized_weight": 0.14285714285714285,
"mode": "live"
},
{
"stage": "cup_out",
"weight": 1,
"normalized_weight": 0.14285714285714285,
"mode": "once"
},
{
"stage": "served",
"weight": 1,
"normalized_weight": 0.14285714285714285,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the coffee scene: one capsule coffee brewed and the filled mug set back on the tray.\n\nRubric stages (weights and modes live in RUBRIC) — the scene's own seven staged flags\n(`CoffeeServiceScene.STAGES`), in the order the appliance's rules impose:\n pod_loaded the capsule seated in the bay pocket (`pod_seated`)\n bay_closed the cover slid shut over the seated capsule (`pod_seated & cover_closed`)\n cup_staged the mug standing on the platform under the spout (`cup_under_spout`)\n started the machine accepted START — it only does with cover closed, pod seated and\n cup staged (the `started` event of the appliance state machine)\n brewed the brew ran to completion: the machine's latched `filled` flag (an abort on\n early open or on a spill never sets it)\n cup_out the filled mug taken off the platform (`filled & ~cup_under_spout`)\n served the filled mug resting upright, settled, on the tray dish — the scene's own\n `success()`\nThe first six are milestones (a loaded pod, a closed bay or a staged cup need not persist once\nthe brew is done and the mug is carried away); the scene latches them itself at substep rate,\nso the grader reads its flags. `brewed` is the machine's irreversible state, `served` is the\nlive end state. Success is the scene's `success()` on the final state.",
"grader_summary": "The mug filled by a completed brew and resting upright, settled, on the serving tray.\n\nLadder: pod loaded 0.143 · bay closed 0.286 · cup staged 0.429 · started 0.571 ·\n brewed 0.714 · cup out 0.857 · served 1.000\n(uniform: one unit per stage of the appliance's chain, the scene's own even credit).",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/puzzle/grader/coffee_service.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/puzzle/grader/coffee_service.py",
"gallery": {
"path": "media/coffee/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/coffee.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "8286701c28128e716c62730200cb08be95ca6772430f5b51672c54d4353fa4ab",
"bytes": 470402
},
"dataset_source": {
"repo_id": "EmbodiedSWE/coffee_franka_joint.jointtarget_60hz_20260907",
"revision": "2450113aeb495c4dd923b72c2a901f1ed84ea884",
"format": "v3.0",
"fps": 60,
"total_episodes": 600,
"camera_order": [
"front_wide",
"wrist"
],
"task_index_scope": "local to this source repository; not a benchmark-global task ID",
"robot_type": "puzzle.coffee.franka.joint",
"selected_episodes": [
0,
300,
599
],
"source_license_metadata": "license: apache-2.0\ntask_categories:\n- robotics\ntags:\n- LeRobot\n- franka\n- puzzle\n- coffee\n- isaac-sim\n- cosigen\n- sim\n- data-gen\nconfigs:\n- config_name: default\n data_files: data/*/*.parquet"
}
},
{
"catalog_number": 15,
"scene_key": "spatula",
"official_scene_id": "puzzle.spatula",
"category": "Puzzle",
"name": "Spatula flip",
"instruction": "Wedge a spatula under bread, flip it, reload it onto the blade, and serve it to a plate.",
"difficulty": "Easy",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "lifted",
"weight": 1,
"normalized_weight": 0.2,
"mode": "once"
},
{
"stage": "wedged",
"weight": 1,
"normalized_weight": 0.2,
"mode": "once"
},
{
"stage": "flipped",
"weight": 1,
"normalized_weight": 0.2,
"mode": "once"
},
{
"stage": "loaded",
"weight": 1,
"normalized_weight": 0.2,
"mode": "once"
},
{
"stage": "served",
"weight": 1,
"normalized_weight": 0.2,
"mode": "live"
}
],
"rubric_variants": {
"serve": [
[
"lifted",
1,
"once"
],
[
"wedged",
1,
"once"
],
[
"loaded",
1,
"once"
],
[
"served",
1
]
],
"flip": [
[
"lifted",
1,
"once"
],
[
"wedged",
1,
"once"
],
[
"flipped_in_pan",
1
]
],
"flip_serve": [
[
"lifted",
1,
"once"
],
[
"wedged",
1,
"once"
],
[
"flipped",
1,
"once"
],
[
"loaded",
1,
"once"
],
[
"served",
1
]
]
},
"grader_description": "Grader for the spatula scene: the bread flipped in the pan and/or served onto the plate on the blade.\n\nThe scene's `cfg.goal` picks the chain (its own `_stage_table`); the rubric follows the goal,\nuniform credit per rung:\n serve lifted -> wedged -> loaded -> served\n flip lifted -> wedged -> flipped_in_pan\n flip_serve lifted -> wedged -> flipped -> loaded (after the flip) -> served\n\nRubric stages (weights and modes live in RUBRIC) — the scene's latched stages, read from its\nsubstep-rate latches (the journey), plus the live destination:\n lifted the blade above the surface by `lift_gate` (`_lifted`) — a milestone\n wedged the bread riding the blade while still down at pan-floor level (`_wedged`)\n flipped the bread inverted by >= `flip_min_deg` AND at rest flat in the pan\n (`_flipped`); a milestone in `flip_serve` (the slice is served afterwards)\n loaded the bread riding the blade above `lift_gate` — the friction carry\n (`_loaded`; in `flip_serve` the scene's `_loaded_pf`, the carry AFTER the\n flip, so a load during the flip lift earns nothing)\n served the scene's `success()` for the serve goals: the bread resting flat on the\n plate now, having ARRIVED ON THE BLADE (`_served`, the contact-history\n clause; `flip_serve` also demands the flip latch)\n flipped_in_pan the scene's `success()` for the flip goal: flipped latched AND the bread\n resting inverted, flat and settled in the pan now\nSuccess is the scene's own goal-dependent `success()` on the final state.",
"grader_summary": "The bread flipped in the pan and/or served flat onto the plate (per the scene's goal), carried there on the blade.\n\nLadder (uniform, one unit per rung of the goal's chain):\n flip_serve: lifted 0.20 · wedged 0.40 · flipped 0.60 · loaded after flip 0.80 · served 1.00\n serve: lifted 0.25 · wedged 0.50 · loaded 0.75 · served 1.00\n flip: lifted 0.333 · wedged 0.667 · flipped in pan 1.000",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/puzzle/grader/spatula_flip_serve.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/puzzle/grader/spatula_flip_serve.py",
"gallery": {
"path": "media/spatula/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/spatula.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "42826d4dc25aaf8cf10f81631940e85aa749e65bdbc21a5ce52a772e6f5359b8",
"bytes": 322286
},
"dataset_source": null
},
{
"catalog_number": 16,
"scene_key": "syringe",
"official_scene_id": "puzzle.syringe",
"category": "Liquid",
"name": "Syringe dosing",
"instruction": "Draw liquid from a reservoir, meter three target doses, and repark the syringe.",
"difficulty": "Hard",
"catalog_embodiments": "Bimanual Franka",
"rubric": [
{
"stage": "drawn",
"weight": 1,
"normalized_weight": 0.2,
"mode": "once"
},
{
"stage": "dosed",
"weight": 3,
"normalized_weight": 0.6,
"mode": "live"
},
{
"stage": "parked",
"weight": 1,
"normalized_weight": 0.2,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the syringe scene: a full draw, one in-band dose in each of three tubes, syringe laid back down.\n\nRubric stages (weights and modes live in RUBRIC):\n drawn the draw: 1 once the scene's `_drawn_ok` latch is set (liquid >= `draw_min` while\n seated on the reservoir); before that, the liquid in the barrel as a linear ramp\n from empty (0) to `draw_min`. A milestone — the liquid is dispensed afterwards\n dosed fraction over the three identical tubes: a tube inside the scene's `dose_band`\n counts 1; below the band, the dose as a linear ramp from 0 to the band's lower\n edge; pushed past the band it counts 0 — wells never un-fill, so an over-dose is\n the scene's irreversible ruin of that tube. Live: an over-push drops a tube\n parked the scene's `success()`: syringe lying at its home spot, on the shelf, settled\n (`parked`) — GATED on the draw + all three doses. The syringe SPAWNS parked, so\n the park only counts once the dosing it concludes is done (else it is free credit)\nSuccess is the scene's own `success()` (drawn AND doses_ok AND parked) on the final state.",
"grader_summary": "A full draw from the reservoir, an in-band dose in each of the three tubes, and the syringe laid back at its spot.\n\nLadder: full draw 0.20 · 1 dose 0.40 · 2 doses 0.60 · 3 doses 0.80 · parked 1.00",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/puzzle/grader/syringe_dosing.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/puzzle/grader/syringe_dosing.py",
"gallery": {
"path": "media/syringe/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/syringe.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "3325f775d9e26b7e993d18a9805190d58bf6bc88cd40c6e8bfdc1add6cb4dacd",
"bytes": 447660
},
"dataset_source": null
},
{
"catalog_number": 17,
"scene_key": "push_shapes",
"official_scene_id": "puzzle.push_shapes",
"category": "Puzzle",
"name": "Push shapes",
"instruction": "Push three blocks onto their matching pads, correcting each block’s orientation along the way.",
"difficulty": "Easy",
"catalog_embodiments": "Unitree G1",
"rubric": [
{
"stage": "yaw_aligned",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
},
{
"stage": "near_pad",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
},
{
"stage": "seated",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the push_shapes scene: the T, X and L blocks each pushed into their own cutout.\n\nRubric stages (weights and modes live in RUBRIC). The three blocks are identical sub-goals,\nso every stage is ONE fraction over the env's three blocks (k/3), never a stage per block:\n yaw_aligned blocks whose orientation is inside the scene's coarse yaw band of their pad\n (`yaw_matched`: cfg.yaw_stage_tolerance_deg) — the pivot push is done\n near_pad blocks whose centre is within cfg.near_xy_tolerance of their own pad centre —\n the translation push has arrived at the cutout\n seated blocks seated in their recess AND settled — the scene's own per-piece success\n gate (`seated() & settled()`: xy within cfg.xy_tolerance, orientation within\n cfg.orientation_tolerance_deg, not lifted, dropped into the recess, at rest)\nAll three are LIVE: a block knocked back out of tolerance loses the credit, exactly as the\nscene's own `success()` is a live conjunction. Every rung is implied by the next (7 mm inside\n30 mm, 7 deg inside 10 deg), so a seated block earns all three. Success is the scene's own\n`success()` — all three blocks seated and settled — read on the final state.",
"grader_summary": "All three blocks pushed into their own cutouts — aligned, dropped in, and at rest.\n\nLadder (3 blocks, 3 equal rungs each, 1/9 per rung): per block yaw aligned +1/9 ·\nnear its pad +1/9 · seated +1/9; 1 block seated 0.333 · 2 blocks seated 0.667 ·\n3 blocks seated 1.00.",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/puzzle/grader/push_shapes.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/puzzle/grader/push_shapes.py",
"gallery": {
"path": "media/push_shapes/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/push_shapes.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "8fff7ad71565b5bec705bf662c1c9c473023c56fd8373493a6d12bc28095c89b",
"bytes": 452044
},
"dataset_source": null
},
{
"catalog_number": 18,
"scene_key": "classify_objects",
"official_scene_id": "puzzle.classify_objects",
"category": "Puzzle",
"name": "Classify objects",
"instruction": "Sort scattered coloured blocks into the zones matching their colours.",
"difficulty": "Easy",
"catalog_embodiments": "Unitree G1",
"rubric": [
{
"stage": "sorted_blocks",
"weight": 1,
"normalized_weight": 1.0,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the classify_objects scene: every coloured block sorted into its matching zone.\n\nRubric stages (weights and modes live in RUBRIC):\n sorted_blocks fraction of the env's blocks correctly sorted — the scene's own\n `sorted_mask()`: centre within cfg.zone_half (xy) of ITS OWN category's\n (jittered) zone centre, resting on the surface (cfg.z_tol) and settled\n (cfg.settle_speed)\nThe blocks are identical pick-and-place sub-goals, so they share this ONE fraction stage\n(k/B). It is LIVE: a block knocked off its zone loses the credit, which is what makes the\nfinal state the only one that counts. No grasp/lift rung: the scene defines no lift\nthreshold, so it is dropped rather than invented. Success is the scene's own `success()`\n— every block sorted — read on the final state.",
"grader_summary": "Every block resting, settled, inside the zone of its own colour.\n\nLadder (k / B blocks sorted; B = len(cfg.manifest), 3 by default): 1 block 0.333 ·\n2 blocks 0.667 · 3 blocks 1.00.",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/puzzle/grader/classify_objects.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/puzzle/grader/classify_objects.py",
"gallery": {
"path": "media/classify_objects/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/classify_objects.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "7693b076de061ca463fe56ee170228546abb3bc154abaf53c9462b53337007d0",
"bytes": 443230
},
"dataset_source": null
},
{
"catalog_number": 19,
"scene_key": "stack_blocks",
"official_scene_id": "puzzle.stack_blocks",
"category": "Puzzle",
"name": "Stack blocks",
"instruction": "Stack scattered blocks into a single aligned tower on a marked pad.",
"difficulty": "Medium",
"catalog_embodiments": "Unitree G1",
"rubric": [
{
"stage": "tower",
"weight": 1,
"normalized_weight": 1.0,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the stack_blocks scene: all blocks stacked into one upright tower on the pad.\n\nRubric stages (weights and modes live in RUBRIC):\n tower contiguous tower height in blocks over cfg.n_blocks — the scene's own\n `tower_height()`: slots 0, 1, 2, ... above the pad centre, each filled by a block\n within cfg.align_tol of the pad axis, at the slot height within cfg.z_tol,\n upright within cfg.upright_max_deg, settled with the whole scene settled\n (cfg.settle_speed), counted up to the first empty slot\nThe cubes are identical sub-goals, so they share this ONE fraction stage (k/n_blocks). It is\nLIVE: a toppled tower loses the credit. No grasp/lift rung: the scene defines no lift\nthreshold, so it is dropped rather than invented. Success is the scene's own `success()` —\nevery slot filled by exactly one block — read on the final state. With n_blocks blocks and\nn_blocks slots, a full-height tower (every slot >= 1 block) is exactly one block per slot,\nso the stage reaches 1 if and only if `success()` holds.",
"grader_summary": "One clean tower on the pad — every slot filled by exactly one upright, settled block.\n\nLadder (k / n_blocks tower height; cfg.n_blocks = 4 by default): 1 block 0.25 ·\n2 blocks 0.50 · 3 blocks 0.75 · 4 blocks 1.00.",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/puzzle/grader/stack_blocks.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/puzzle/grader/stack_blocks.py",
"gallery": {
"path": "media/stack_blocks/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/stack_blocks.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "7827adcc799e671f6e55acfdb4bfc620b538751a8bbcaf0b9008aae2eb7c2c5a",
"bytes": 530682
},
"dataset_source": null
},
{
"catalog_number": 20,
"scene_key": "tshirt",
"official_scene_id": "deformable.tshirt",
"category": "Deformable",
"name": "T-shirt folding",
"instruction": "Fold a T-shirt into a compact, flat bundle.",
"difficulty": "Medium",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "sleeves",
"weight": 2,
"normalized_weight": 0.4,
"mode": "live"
},
{
"stage": "hem",
"weight": 1,
"normalized_weight": 0.2,
"mode": "live"
},
{
"stage": "folded",
"weight": 1,
"normalized_weight": 0.2,
"mode": "live"
},
{
"stage": "plausible",
"weight": 1,
"normalized_weight": 0.2,
"mode": "final"
}
],
"rubric_variants": null,
"grader_description": "Grader for the tshirt scene: the shirt folded flat — sleeves to the center line, hem up over the collar.\n\nRubric stages (weights and modes live in RUBRIC):\n sleeves fraction of the two sleeve folds done — for each sleeve, the linear progress of\n its tip particle (the max-x / min-x cloth particle at the graded reset) from its\n reset x to the shirt's center line (x = 0, the table's axis); two identical\n sub-goals, so ONE stage worth two rungs, value k/2\n hem the bottom hem's edge particle (max-y at reset) carried from its reset y to the\n collar's reset y (min-y at reset): linear, 1 at or past the collar\n folded the scene's own success: footprint (x-extent · y-extent of the cloth) at or\n below `cfg.fold_footprint_max`\n plausible the VLM plausibility gate on the rendered final frame (final rung): a neat flat\n compact bundle on the table — rejects crumpled/balled cloth, cloth hanging off\n the table, cloth still spread out\nEvery stage is per env. Success = the scene's `success()` (footprint gate) on the final state\nAND the plausibility gate — the gate can only take credit away.\n\nNot measurable from state (hence the gate): whether the small footprint is a fold or a crumple,\nand whether the cloth rests on the table rather than hanging off it — the scene has no such\npredicates and inventing tolerances is not allowed.",
"grader_summary": "The T-shirt folded into a flat compact bundle on the table (footprint gate), judged plausible.\n\nLadder: 1 sleeve 0.20 · 2 sleeves 0.40 · hem over collar 0.60 · folded (footprint) 0.80 · gate 1.00",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/deformable/grader/tshirt.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/deformable/grader/tshirt.py",
"gallery": {
"path": "media/tshirt/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/tshirt.mp4",
"website_playback_rate": 2.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "833f6a1dd107ca3404eec7324c2e8fef5d422e317698b852eb4f765c71982c29",
"bytes": 699184
},
"dataset_source": null
},
{
"catalog_number": 21,
"scene_key": "latte",
"official_scene_id": "deformable.latte",
"category": "Liquid",
"name": "Latte pouring",
"instruction": "Coordinate two arms to pour milk into a carried coffee mug without excessive spilling.",
"difficulty": "Hard",
"catalog_embodiments": "Bimanual Franka",
"rubric": [
{
"stage": "pitcher_grasped",
"weight": 1,
"normalized_weight": 0.25,
"mode": "once"
},
{
"stage": "transferred",
"weight": 1,
"normalized_weight": 0.25,
"mode": "live"
},
{
"stage": "poured",
"weight": 1,
"normalized_weight": 0.25,
"mode": "live"
},
{
"stage": "plausible",
"weight": 1,
"normalized_weight": 0.25,
"mode": "final"
}
],
"rubric_variants": null,
"grader_description": "Grader for the latte scene: milk poured into the coffee without emptying the pitcher or spilling.\n\nRubric stages (weights and modes live in RUBRIC):\n pitcher_grasped the pitcher taken — the scene's own auto-weld grasp contract engaged on the\n pitcher's handle bar by either hand (`_auto_state` of `weld_pitcher` /\n `weld_pitcher_l`); a milestone, credit kept after the release\n transferred linear ramp of the scene's `transfer_fraction()` (milk particles inside the\n mug) from 0 to `cfg.success_transfer_min`\n poured the scene's own success: all four gates at once — milk transferred\n (>= success_transfer_min), milk kept in the pitcher (>= success_kept_min),\n coffee retained (>= success_retention_min), nothing spilled\n (<= success_spilled_max). The three guards are sanity predicates on the done\n rung (a live stage that is true from the first step would score a run that\n did nothing), not weighted rungs of their own\n plausible the VLM plausibility gate on the rendered final frame (final rung): two\n upright vessels resting on the table, coffee with milk poured in — rejects\n the pitcher inside/on the mug, tipped vessels, vessels held mid-air or off\n the table, liquid pooled on the table\nEvery stage is per env (the MPM scene is single-env; scalars broadcast). Success = the scene's\n`success()` on the final state AND the plausibility gate.\n\nNot measurable from state (hence the gate): a pitcher dunked into the mug passes every particle\nmetric; \"upright and resting on the table\" has no scene predicate, and inventing a tilt/height\ntolerance is not allowed.",
"grader_summary": "Milk poured into the coffee mug — the scene's four gates met — with both vessels left upright on the table.\n\nLadder: pitcher grasped 0.25 · milk transferred 0.50 · poured (4 gates) 0.75 · gate 1.00",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/deformable/grader/latte.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/deformable/grader/latte.py",
"gallery": {
"path": "media/latte/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/latte.mp4",
"website_playback_rate": 3.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "0c2d5974441e189c3a1764f553509ebe868b7c41e66558dc04f6436909281003",
"bytes": 811918
},
"dataset_source": null
},
{
"catalog_number": 22,
"scene_key": "dumpling",
"official_scene_id": "deformable.dumpling",
"category": "Deformable",
"name": "Dough rolling",
"instruction": "Roll a ball of dough flat into a round dumpling wrapper with a rolling pin.",
"difficulty": "Hard",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "flattened",
"weight": 1,
"normalized_weight": 0.25,
"mode": "live"
},
{
"stage": "spread",
"weight": 1,
"normalized_weight": 0.25,
"mode": "live"
},
{
"stage": "rolled",
"weight": 1,
"normalized_weight": 0.25,
"mode": "live"
},
{
"stage": "plausible",
"weight": 1,
"normalized_weight": 0.25,
"mode": "final"
}
],
"rubric_variants": null,
"grader_description": "Grader for the dumpling scene: the dough ball rolled out into a thin, wide, compact wrapper.\n\nRubric stages (weights and modes live in RUBRIC):\n flattened linear ramp of the scene's `flatten_height()` (q95 dough height above the table)\n from its value at the graded reset (the ball, ~0.044 m) down to\n `cfg.flatten_h95_max`\n spread linear ramp of the scene's `spread_radius()` (q90 radius about the dough centroid)\n from its reset value (~0.020 m) up to `cfg.spread_r90_min`\n rolled the scene's own success: thin (flatten_h95_max) AND wide (spread_r90_min) AND the\n guards — compact (`final_extent() <= extent_max`), conserved\n (`conservation() >= conserve_min`), nothing punched through the table\n (`min_particle_z() >= surface_z - floor_z_slack`). The guards are sanity predicates\n on the done rung, not weighted rungs (they hold from the first step)\n plausible the VLM plausibility gate on the rendered final frame (final rung): one round,\n flat, even wrapper on the board with the pin set aside — rejects splats and\n irregular shapes, torn/multiple pieces, dough on the pin or the pin in the dough,\n dough off the board\nEvery stage is per env (the MPM scene is single-env). Success = the scene's `success()` on the\nfinal state AND the plausibility gate.\n\nNo pin-grasp rung: the scene's success does not involve the pin (dough flattened by any means\npasses), so a grasp milestone could leave progress below 1 on an official success.\nNot measurable from state (hence the gate): roundness/evenness of the sheet and whether the pin\nis set aside — thin + wide + compact is also satisfied by a splat hammered with the pin's end.",
"grader_summary": "The dough rolled out into a thin, wide, compact wrapper on the board — the scene's gates met — judged plausible.\n\nLadder: flattened 0.25 · spread 0.50 · rolled (scene gates) 0.75 · gate 1.00",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/deformable/grader/dumpling.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/deformable/grader/dumpling.py",
"gallery": {
"path": "media/dumpling/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/dumpling.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "5add18c9c44c41b506eedae251f2305ec51a55798f21de9fa7f35138f74c7077",
"bytes": 577626
},
"dataset_source": null
},
{
"catalog_number": 23,
"scene_key": "shoe_knot",
"official_scene_id": "deformable.shoe_knot",
"category": "Deformable",
"name": "Shoelace half-knot",
"instruction": "Tie a half knot from the two shoelaces of a sneaker.",
"difficulty": "Hard",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "wound",
"weight": 1,
"normalized_weight": 0.25,
"mode": "live"
},
{
"stage": "snug",
"weight": 1,
"normalized_weight": 0.25,
"mode": "live"
},
{
"stage": "tied",
"weight": 1,
"normalized_weight": 0.25,
"mode": "live"
},
{
"stage": "plausible",
"weight": 1,
"normalized_weight": 0.25,
"mode": "final"
}
],
"rubric_variants": null,
"grader_description": "Grader for the knot scene: a self-holding half knot tied from the two laces, seated on the tongue.\n\nRubric stages (weights and modes live in RUBRIC):\n wound linear ramp of the scene's `knot_winding()` (lace 1's winding around lace 2 on the\n knot sections, degrees) from 0 to `cfg.winding_min_deg`\n snug linear ramp of the scene's `knot_contacts()` (cross-lace contact pairs on the knot\n sections) from 0 to `cfg.contacts_min`\n tied the scene's own `success(world)`: winding >= winding_min_deg AND contacts >=\n contacts_min AND the knot (closest cross-lace pair) seated at task-local\n z < `cfg.knot_z_max`. The seat gate is a sanity predicate on this done rung, not a\n rung of its own: with the laces at rest the closest cross-lace pair already sits\n at z ~ 0.12 < 0.155, so a standalone height rung would score a run that did nothing\n plausible the VLM plausibility gate on the rendered final frame (final rung): a single half\n knot on the tongue, shoe upright and in place — rejects laces wrapped around the\n shoe, shoe moved/flipped, laces merely crossed or lying loose, ends tied around\n anything else\nThe scene's metrics are numpy, per world: every stage loops over the worlds and returns a\n`(num_envs,)` tensor. Success = the scene's `success(world)` on the final state AND the\nplausibility gate.\n\nNot measurable from state (hence the gate): the scene exposes no slack/hold check (the smoke's\ncheck window is smoke logic, not a scene predicate) and no \"knot is around lace 2, not the shoe\"\ntest — winding + contacts are also satisfied by laces wound around the shoe body.",
"grader_summary": "A self-holding half knot tied from the two laces and seated on the tongue — the scene's gates met — judged plausible.\n\nLadder: wound 0.25 · snug 0.50 · tied (scene gates) 0.75 · gate 1.00",
"success_source": "def check_success(self):\n import torch\n\n return torch.tensor([bool(self.scene.success(w)) for w in self._worlds()])",
"grader_source_path": "robobench/suites/deformable/grader/shoe_knot.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/deformable/grader/shoe_knot.py",
"gallery": null,
"dataset_source": null
},
{
"catalog_number": 24,
"scene_key": "slice",
"official_scene_id": "cutting.slice",
"category": "Cutting",
"name": "Banana slicing",
"instruction": "Pick up a chef knife from its rest and slice a carrot (or a banana) into a target number of pieces on a chopping board.",
"difficulty": "Easy",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "knife_taken",
"weight": 1,
"normalized_weight": 0.1111111111111111,
"mode": "once"
},
{
"stage": "planes_cut",
"weight": 7,
"normalized_weight": 0.7777777777777778,
"mode": "live"
},
{
"stage": "sliced",
"weight": 1,
"normalized_weight": 0.1111111111111111,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the slice scene: the food cut into its target number of pieces, all still on the board.\n\nRubric stages (weights and modes live in RUBRIC):\n knife_taken the knife has been lifted clear of its rest — its root raised by at least\n cfg.rest_rail_h (the rails' height) above where it lay at the graded reset, so\n the edge is above the rail tops and the notches no longer hold it; a milestone\n (\"once\"), credit kept after the knife is set down again\n planes_cut fraction of the cuts the target needs that have been made — released planes\n over (target - 1), a linear ramp from the welded food (0) to the target piece\n count (1), target = cfg.target_pieces or every piece. The scored planes are\n identical sub-goals: one stage, one fraction, each plane worth the same\n sliced the scene's own done test: piece count (connected components of the live weld\n chain, `pieces_count()`) reaches the target AND every piece is still on the\n chopping board — inside the board's footprint (cfg.board_size) and above its\n top. \"On the board\" is a GATE on this stage, not a stage of its own: it is\n true at reset and only means anything once the food is cut\nSuccess = `sliced` read when the delivery finishes.",
"grader_summary": "Food sliced into the target number of pieces, every piece still on the board.\n\nLadder (7 scored planes — the carrot / banana assets — target = all pieces): knife taken\n1/9 (0.111) · each plane cut +1/9 (7 planes -> 8/9 = 0.889) · sliced with every piece on\nthe board 1.00.",
"success_source": "def check_success(self):\n return self._sliced()",
"grader_source_path": "robobench/suites/cutting/grader/slice_food.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/cutting/grader/slice_food.py",
"gallery": {
"path": "media/slice/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/slice_banana.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "26ce8e19831ee420a2bb8dd74e46d90ac1e8e6b832e6c65fbc777a216b8fb547",
"bytes": 399060
},
"dataset_source": {
"repo_id": "EmbodiedSWE/cutting_slice_franka_diffik.jointtarget_15hz_train_20260919",
"revision": "0b89ef8fd191fd09815e7edb8e5fcb467f95e494",
"format": "v3.0",
"fps": 15,
"total_episodes": 4198,
"camera_order": [
"front",
"wrist"
],
"task_index_scope": "local to this source repository; not a benchmark-global task ID",
"robot_type": "cutting.slice_banana.franka.diff_ik",
"selected_episodes": [
0,
1055,
3664
],
"source_license_metadata": "None"
}
},
{
"catalog_number": 25,
"scene_key": "dice",
"official_scene_id": "cutting.dice",
"category": "Cutting",
"name": "Food dicing",
"instruction": "Dice a tomato (or a potato) into a 3×3 grid of pieces, cutting along two plane families with the knife re-yawed between them.",
"difficulty": "Medium",
"catalog_embodiments": "Franka / xArm7 / Gen3",
"rubric": [
{
"stage": "knife_taken",
"weight": 1,
"normalized_weight": 0.07142857142857142,
"mode": "once"
},
{
"stage": "pairs_cut",
"weight": 12,
"normalized_weight": 0.8571428571428571,
"mode": "live"
},
{
"stage": "diced",
"weight": 1,
"normalized_weight": 0.07142857142857142,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the dice scene: the food diced into its target number of pieces, all still on the board.\n\nRubric stages (weights and modes live in RUBRIC):\n knife_taken the knife has been lifted off its rest — its root raised by the rest's rail\n height (`cfg.rest_rail_h`), the lift that frees the plate from the notches; a\n milestone, credit kept after the knife is put down\n pairs_cut fraction of the grid's weld PAIRS released (`scene.pair_cut`, the dice gate's\n own truth). The x-planes and y-planes are identical sub-goals — one stage, one\n fraction, each pair worth the same\n diced the scene's own done test: connected components of the live weld graph reach\n the target piece count AND every piece is still on the board (footprint from\n `cfg.board_size`, resting above the board top). \"On the board\" is a GATE on\n this stage, not a stage of its own — it is true at reset and only means\n anything once the food is diced\nSuccess = `diced` read on the final state.",
"grader_summary": "Food diced into the target number of pieces, every piece still on the board.\n\nLadder: knife taken 1/14 (0.071) · each weld pair released +1/14 (12 pairs -> 13/14 =\n0.929) · diced with every piece on the board 1.00",
"success_source": "def check_success(self):\n return self._diced()",
"grader_source_path": "robobench/suites/cutting/grader/dice_food.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/cutting/grader/dice_food.py",
"gallery": null,
"dataset_source": null
},
{
"catalog_number": 26,
"scene_key": "fruit_delivery",
"official_scene_id": "locomanip.fruit_delivery",
"category": "Locomanip",
"name": "Fruit delivery",
"instruction": "Pick up every fruit on a long table and carry it around the table to a plate beyond arm reach.",
"difficulty": "Medium",
"catalog_embodiments": "Unitree G1",
"rubric": [
{
"stage": "lifted",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "once"
},
{
"stage": "carried",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "once"
},
{
"stage": "placed",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the fruit delivery scene: every fruit carried around the table and set on the far plate.\n\nRubric stages (weights and modes live in RUBRIC), each a fraction over the fruits PRESENT this\nepisode (the scene's `present` mask; identical fruits share one stage):\n lifted fraction of fruits ever raised clear of the table — the fruit's origin (the produce\n assets sit with their origin at the base, on `cfg.surface_z`) above the plate's rim\n height (`cfg.plate_rim_z`), the height every fruit must clear to enter the dish; a\n milestone, credit kept after the fruit is set down\n carried mean transit progress: for each fruit, a linear ramp of its radial xy distance from\n the plate's counted centre (the scene's own body-frame measure, see `_on_plate`) from\n where it lay at the graded reset (0) to the counted footprint edge\n `cfg.radius_frac * cfg.plate_radius` (1); a milestone, so a fruit that arrives and is\n released keeps the credit\n placed the scene's own `fruits_placed()` — on the plate AND settled — as a fraction of the\n present fruits, GATED by the scene's `nonfruits_on_plate()` being empty: a pumpkin or\n onion on the plate fails the task, so it zeroes the done stage rather than being a\n weighted stage of its own (it is trivially \"true\" at reset)\nSuccess is the scene's own `success()`: every present fruit placed and nothing that is not a fruit\non the plate, read on the final state.",
"grader_summary": "Every fruit carried around the table and resting on the far-side plate, nothing else put on it.\n\nLadder (one fruit): lifted 0.333 · carried 0.667 · placed 1.00\nWith N fruits every rung is the fraction over the N present fruits, so k fruits fully\ndelivered score k/N.",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/locomanip/grader/fruit_delivery.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/locomanip/grader/fruit_delivery.py",
"gallery": {
"path": "media/fruit_delivery/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/fruit_delivery.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "e6d35cb7c44c87f2c5c1fbefca70977021f2e96429da66646c3500ad3c5bbaea",
"bytes": 328419
},
"dataset_source": null
},
{
"catalog_number": 27,
"scene_key": "box_to_bin",
"official_scene_id": "locomanip.box_to_bin",
"category": "Locomanip",
"name": "Box to bin",
"instruction": "Pick a cardboard box off a shelf and carry it to a sorting bin on a table across the room.",
"difficulty": "Medium",
"catalog_embodiments": "Unitree G1",
"rubric": [
{
"stage": "picked",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "once"
},
{
"stage": "carried",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "once"
},
{
"stage": "in_bin",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the box-to-bin scene: the box taken off the shelf, walked around the corner, and left in the bin.\n\nRubric stages (weights and modes live in RUBRIC) — the scene's own chain lifted -> transited ->\nplaced, one rung each:\n picked the scene's `lifted` latch: the box got clear of its shelf board, either raised\n `cfg.lift_h` above its resting height or seen held up in the apron in front of the\n shelf face (both pick styles the open shelving affords); a milestone\n carried the scene's own `carried_fraction()` — the fraction of the start-to-bin xy distance\n the box has closed, a linear ramp — reading 1 once the scene's `transited` latch has\n fired (its own \"the transit is done\" mark, which `success()` requires); a milestone,\n so a box that arrives and is released keeps the credit\n in_bin the scene's own `placed()` — fully inside the bin's inset interior bands, below the\n rim-excluding height cap, and settled; live, so it dips if the box leaves the bin\nSuccess is the scene's own `success()` — `lifted & transited & placed()` — read on the final\nstate; the three rungs are exactly its three conjuncts, so progress reads 1.0 iff it holds.",
"grader_summary": "The cardboard box picked off the shelf, carried around the corner, and resting inside the sorting bin.\n\nLadder: picked 0.333 · carried 0.667 (linear in the distance closed) · in bin 1.00",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/locomanip/grader/box_to_bin.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/locomanip/grader/box_to_bin.py",
"gallery": {
"path": "media/box_to_bin/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/box_to_bin.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "7819414f4155c0e5f90b9f25ab4aa07dbfaacb05f10872e96796c7cf3417bb3a",
"bytes": 175299
},
"dataset_source": null
},
{
"catalog_number": 28,
"scene_key": "wheel_carry",
"official_scene_id": "locomanip.wheel_carry",
"category": "Locomanip",
"name": "Wheel carry",
"instruction": "Pick a steering wheel off one packing table and carry it to a basket on another, three metres away.",
"difficulty": "Medium",
"catalog_embodiments": "Unitree G1",
"rubric": [
{
"stage": "picked",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "once"
},
{
"stage": "carried",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "once"
},
{
"stage": "in_basket",
"weight": 1,
"normalized_weight": 0.3333333333333333,
"mode": "live"
}
],
"rubric_variants": null,
"grader_description": "Grader for the wheel carry scene: the wheel picked up here, walked over there, and left in the basket.\n\nRubric stages (weights and modes live in RUBRIC) — the scene's own chain lifted -> transited ->\nplaced, one rung each:\n picked the scene's `lifted` latch: the wheel's origin raised `cfg.lift_h` above the pick\n table top; a milestone, credit kept after it is put down\n carried the scene's own `carried_fraction()` — the fraction of `cfg.carry_dx` the wheel\n has travelled along +x from the pick station, a linear ramp — reading 1 once the\n scene's `transited` latch has fired (its own \"the transit is done\" mark, which\n `success()` requires); a milestone, so a wheel that arrives and is released keeps\n the credit\n in_basket the scene's own `placed()` — inside the basket's target bands, below the height\n cap, and settled; live, so it dips if the wheel leaves the basket\nSuccess is the scene's own `success()` — `lifted & transited & placed()` — read on the final\nstate; the three rungs are exactly its three conjuncts, so progress reads 1.0 iff it holds.",
"grader_summary": "The steering wheel lifted off the near table, carried across, and resting inside the far basket.\n\nLadder: picked 0.333 · carried 0.667 (linear in the distance closed) · in basket 1.00",
"success_source": "def check_success(self):\n return self.scene.success()",
"grader_source_path": "robobench/suites/locomanip/grader/wheel_carry.py",
"grader_url": "https://github.com/EmbodiedSWE/EmbodiedSWE/blob/433d0ceedc2c5acf58551f4b3502b2cd1c7e8f1b/robobench/suites/locomanip/grader/wheel_carry.py",
"gallery": {
"path": "media/wheel_carry/website_demo.mp4",
"url": "https://embodiedswe.github.io/assets/video/wheel_carry.mp4",
"website_playback_rate": 1.0,
"camera_label": "WEBSITE RENDER VIEW",
"sha256": "79805b9c731ab40666c5c1abe55203257880e0d15d72656125a0f5dd1bd6514a",
"bytes": 367523
},
"dataset_source": null
}
] |