Datasets:
This dataset was created using LeRobot.
Dataset Description
license: apache-2.0 tags: [sciedu-bench, simulation, mujoco, openarm, failures]
SciEdu-Bench 시뮬 실패 예시 (MuJoCo, 장면 props-9)
학습 데이터가 아니다. 실패 예시다. 구간 성공 판정기(단계가 끝났을 때 "됐다/안 됐다"를 고르는 모델)가 실패 장면을 보도록 모은 것이다. 이 편들로 정책을 학습하면 실패를 따라 하게 된다.
어디서 왔나
SciEdu-Bench 학습 시연(leapshared/sciedu_bench_v2_sim_*)을 지을 때 생성기는 무작위로 흩뜨린 시작마다 IK 코드 시연을 돌려 시뮬 판정을
지난 판만 남겼다. 일부러 실패하게 만든 것이 아니다. 같은 코드가 어떤 시작에서는 판정을 넘지 못했다: 비커를 밀거나, 판을 덜 담그거나,
막대가 검전기에 닿거나, 긴 과제의 한 단계에서 끊겼다. 그 판(계획 파일에 시작과 판정 숫자만 남은 47편)을 같은 시작·같은
생성기로 다시 돌려 그림을 담았다. 다시 돌린 판이 기록과 다르면(틱 수·판정 숫자·끊긴 단계·실패 꼴) 담지 않았다: 없음. 기록이 0틱인 판 1편은 그 시작에서 코드 시연을 짓지 못한 것이라 담을 그림이 없다.
- 편 46, 프레임 68871, 30 fps, 16값 절대 관절 도, 카메라 셋 480×640(학습 데이터와 같은 꼴, LeRobot v3.0).
- 장면 판
props-9(SciEdu-Bench 꾸러미 0.9.0-draft의 순위 장면). 광물 편은 광물 규격 장면(mineral-3) 위다. - 꾸러미에서 걷어 낸 과제의 실패는 담지 않았다: 375편(
redox_dip8,redox_home1,redox_lift5,redox_move_blue21,redox_move_blue_look22,redox_point42,redox_put_left10,redox_script266). - 긴 과제는 처음 진 단계와 그 뒤 기다림까지만 담았다(9편). 그 뒤는 망가진 장면 위의 판정이라 라벨로 쓰지 않는다. 다시 돌리기는
끝까지 했다(물리·판정이 기록과 같은지 본 뒤 자름). 자르기 전 길이는 곁 파일
frames_full(합 94166).
편마다(곁 파일 sim_episodes.jsonl)
success false, verdict "fail", p_ok null(시뮬 판정은 확률이 없다), fail_kind(아래 표), fail_frame(실패가 드러난 프레임: 가드
실패는 처음 넘은 프레임, 나머지는 진 단계의 끝 프레임, fail_frame_rule), steps(단계마다 step{task, number, position, sentence},
first_frame·end_frame, t_end 초, verdict ok/fail, fail_kind, end_values: 판정 숫자·문턱·여유·문턱 ±15% 띠 안인지 in_band),
judge(시뮬 판정 코드·규격 판·장면 판·코드 커밋), info(판정 줄 그대로), start(시작 상태), plan(원 계획 파일과 줄).
꼴은 판정 계약 초안(plan/segment-success.md 6.3)을 따른다. 판정 문턱 ±15% 띠 안에서 진 편이 2편이다: 그림으로는 성공과 거의 같아
"모름"으로 두기를 권한다(같은 문서 4.2).
실패 꼴
| 꼴 | 뜻 | 편 |
|---|---|---|
goal_not_reached |
상한 틱까지 목표에 못 미침 | 14 |
cup_disturbed |
비커를 12°보다 기울이거나 2 cm보다 밂 | 8 |
chain_broken@1 |
긴 과제의 단계가 끊김 | 4 |
chain_broken@7 |
긴 과제의 단계가 끊김 | 4 |
lost_plate |
쥔 것을 놓침 | 3 |
prop_disturbed |
소품(받침)을 밂 | 3 |
chain_broken@10 |
긴 과제의 단계가 끊김 | 3 |
chain_broken@14 |
긴 과제의 단계가 끊김 | 2 |
chain_broken@3 |
긴 과제의 단계가 끊김 | 2 |
rod_touched_electroscope |
막대가 검전기에 닿음 | 1 |
never_grasped |
쥐지 못함 | 1 |
chain_broken@11 |
긴 과제의 단계가 끊김 | 1 |
생성기가 남긴 까닭(광물 판정의 낱말 따위, 옛 산화환원 줄은 없어 위 꼴): cup_disturbed 8, goal_not_reached 8, not_drawn 6, chain_broken@1 4, chain_broken@7 4, lost_plate 3, rack_pushed 3, chain_broken@10 3, chain_broken@14 2, chain_broken@3 2, rod_touched_electroscope 1, never_grasped 1, chain_broken@11 1.
과제
| 과제 | 편 |
|---|---|
mineral_pick_dropper |
9 |
redox_steps_script |
9 |
mineral_draw |
6 |
mineral_round |
6 |
redox_steps_lift |
6 |
redox_steps_raise |
3 |
mineral_round_swap |
2 |
redox_steps_dip |
2 |
electroscope_rod_put |
1 |
leaf_script |
1 |
redox_steps_put_left |
1 |
English
Failure examples, not training demonstrations: IK code demos of SciEdu-Bench that the simulator's judge failed while the
training data was being generated from randomized starts (not failures made on purpose), played again from the same start with
the same generator and recorded with the three cameras. Lines whose re-run differed from the record were left out (없음). 1 line(s) recorded 0 ticks (the code demo could not be built at that start): no pictures.
So were the failures of tasks the package retired (375, the old redox board).
Long tasks are cut after their first failed step and its wait. Each episode's notes (sim_episodes.jsonl) carry the verdict
contract fields (step, t_end, verdict fail, p_ok null, fail_kind, end_values with thresholds, judge) and the frame its failure
became plain. Scene props-9, 46 episodes, 68871 frames. License apache-2.0.
- Homepage: [More Information Needed]
- Paper: [More Information Needed]
- License: apache-2.0
Dataset Structure
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"action": {
"dtype": "float32",
"shape": [
16
],
"names": [
"left_joint_1.pos",
"left_joint_2.pos",
"left_joint_3.pos",
"left_joint_4.pos",
"left_joint_5.pos",
"left_joint_6.pos",
"left_joint_7.pos",
"left_gripper.pos",
"right_joint_1.pos",
"right_joint_2.pos",
"right_joint_3.pos",
"right_joint_4.pos",
"right_joint_5.pos",
"right_joint_6.pos",
"right_joint_7.pos",
"right_gripper.pos"
]
},
"observation.state": {
"dtype": "float32",
"shape": [
16
],
"names": [
"left_joint_1.pos",
"left_joint_2.pos",
"left_joint_3.pos",
"left_joint_4.pos",
"left_joint_5.pos",
"left_joint_6.pos",
"left_joint_7.pos",
"left_gripper.pos",
"right_joint_1.pos",
"right_joint_2.pos",
"right_joint_3.pos",
"right_joint_4.pos",
"right_joint_5.pos",
"right_joint_6.pos",
"right_joint_7.pos",
"right_gripper.pos"
]
},
"observation.images.follower_d455f": {
"dtype": "video",
"shape": [
480,
640,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 480,
"video.width": 640,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.fps": 30,
"video.channels": 3,
"has_audio": false,
"video.g": 2,
"video.crf": 30,
"video.preset": 12,
"video.fast_decode": 0,
"video.video_backend": "pyav",
"video.extra_options": {},
"is_depth_map": false
}
},
"observation.images.left_wrist": {
"dtype": "video",
"shape": [
480,
640,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 480,
"video.width": 640,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.fps": 30,
"video.channels": 3,
"has_audio": false,
"video.g": 2,
"video.crf": 30,
"video.preset": 12,
"video.fast_decode": 0,
"video.video_backend": "pyav",
"video.extra_options": {},
"is_depth_map": false
}
},
"observation.images.right_wrist": {
"dtype": "video",
"shape": [
480,
640,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 480,
"video.width": 640,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.fps": 30,
"video.channels": 3,
"has_audio": false,
"video.g": 2,
"video.crf": 30,
"video.preset": 12,
"video.fast_decode": 0,
"video.video_backend": "pyav",
"video.extra_options": {},
"is_depth_map": false
}
},
"timestamp": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"frame_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"episode_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"task_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
}
},
"total_episodes": 46,
"total_frames": 68871,
"total_tasks": 22,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path": "videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4",
"robot_type": "openarm_sciedu_sim",
"splits": {
"train": "0:46"
}
}
Citation
BibTeX:
[More Information Needed]
- Downloads last month
- 70