devorah-ai-2016 commited on
Commit
ec9601d
Β·
1 Parent(s): 30060a0

Release MicroDuck Locomotion Dataset V1.0 Zero-Slip Baseline (250,000 frames)

Browse files
README.md ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-sa-4.0
3
+ task_categories:
4
+ - robotics
5
+ - reinforcement-learning
6
+ tags:
7
+ - bipedal-robotics
8
+ - locomotion
9
+ - sim-to-real
10
+ - mujoco
11
+ - zero-slip
12
+ - microduck
13
+ size_categories:
14
+ - 100K<n<1M
15
+ configs:
16
+ - config_name: default
17
+ data_files:
18
+ - split: train
19
+ path: data_part_*.csv
20
+ ---
21
+
22
+ # πŸ¦† MicroDuck Locomotion Dataset V1.0: Zero-Slip Flat Ground Baseline
23
+
24
+ **Publisher**: `devorah-ai-2026` / `devorahai2026`
25
+ **License**: `Commercial / CC-BY-SA 4.0`
26
+ **Task**: Bipedal Locomotion, Reinforcement Learning, Sim-to-Real
27
+ **Hardware Platform**: Antigravity 14-DOF MicroDuck Biped Robot
28
+
29
+ ---
30
+
31
+ ## πŸ“Œ Overview (κ°œμš”)
32
+ λ³Έ 데이터셋은 **14-DOF μ†Œν˜• 2μ‘± 보행 λ‘œλ΄‡(MicroDuck)**의 μ™„λ²½ν•œ **평면 보행(Flat-ground walking)** ꢀ적 및 μ œμ–΄ 둜그λ₯Ό 담은 κ³ ν’ˆμ§ˆ ν•©μ„± 데이터(Synthetic Data)μž…λ‹ˆλ‹€. κΈ°μ‘΄ κ°•ν™”ν•™μŠ΅ 기반 2μ‘± 보행 μ—μ΄μ „νŠΈλ“€μ΄ ν”νžˆ κ²ͺλŠ” **λ°œλ°”λ‹₯ λ―Έλ„λŸ¬μ§(Ice-Skating)** 및 **제자리걸음(Marching in place)** 문제λ₯Ό κ³΅ν•™μ μœΌλ‘œ μ™„λ²½νžˆ κ·Ήλ³΅ν•œ μ΅œμƒκΈ‰ 베이슀라인 λͺ¨λΈμž…λ‹ˆλ‹€. ν–₯ν›„ λ³΅μž‘ν•œ μ§€ν˜• ν™˜κ²½μ΄λ‚˜ λΉ„μ „(Vision) 기반 μ‘°ν–₯ ν•™μŠ΅μ„ μœ„ν•œ ν†΅μ œλœ λŒ€μ‘°κ΅°(Control Group)으둜 ν™œμš©ν•˜κΈ°μ— μ΅œμ ν™”λ˜μ–΄ μžˆμŠ΅λ‹ˆλ‹€.
33
+
34
+ ---
35
+
36
+ ## πŸš€ Key Features (핡심 기술 사양)
37
+
38
+ * **Zero-Slip Planted Stance**: μ§€λ©΄ 접촉 μƒνƒœ($\text{contact} > 0.5$)μ—μ„œ λ””λ”€λ°œμ˜ μˆ˜ν‰ 속도가 **$0.000 \pm 0.002\text{ m/s}$**둜 μˆ˜λ ΄ν•˜μ—¬ μ™„λ²½ν•œ μ§€λ©΄ λ°€μ°© 보행을 κ΅¬ν˜„ν•©λ‹ˆλ‹€.
39
+ * **Controlled Falling Momentum**: 상체 ν”ΌμΉ˜(Trunk Pitch)λ₯Ό **λ―Έμ„Έ μ „λ°© λ°”μ΄μ–΄μŠ€($-3.0^\circ$)**둜 μ œμ–΄ν•˜μ—¬ 쀑λ ₯에 μ˜ν•œ μžμ—°μŠ€λŸ¬μš΄ μ „μ§„ λͺ¨λ©˜ν…€μ„ ν˜•μ„±ν•˜κ³ , λ””λ”€λ°œμ˜ **κ°•λ ₯ν•œ ν›„λ°© ν‚₯(Push-off) μΆ”λ ₯($+0.22\text{ rad}$)**을 데이터에 λ‹΄μ•˜μŠ΅λ‹ˆλ‹€.
40
+ * **High Energy Efficiency**: μ§€λ©΄ 긁음(Ground Scraping)κ³Ό 헛바퀴 토크 손싀을 μ œκ±°ν•˜μ—¬ μ‹€μ œ λͺ¨ν„° 이식 μ‹œ λ°œμ—΄ 및 배터리 μ†Œλͺ¨λ₯Ό λŒ€ν­ μ ˆκ°ν•  수 μžˆλŠ” κΆ€μ μž…λ‹ˆλ‹€.
41
+ * **Swarm Tested**: 10기의 닀쀑 μ—μ΄μ „νŠΈ ν™˜κ²½μ—μ„œ **500회 연속 μ—ν”Όμ†Œλ“œ(250,000 ν”„λ ˆμž„)** λ™μ•ˆ μ—‰ν‚΄μ΄λ‚˜ 전볡 없이 μ•ˆμ •μ μœΌλ‘œ μ£Όν–‰ν•œ κ²€μ¦λœ λ°μ΄ν„°μž…λ‹ˆλ‹€.
42
+
43
+ ---
44
+
45
+ ## πŸ“Š Dataset Structure (데이터 ꡬ쑰)
46
+
47
+ 데이터셋은 500회 연속 μ£Όν–‰ ν…ŒμŠ€νŠΈ λ²€μΉ˜μ—μ„œ μΆ”μΆœλœ νƒ€μž„μŠ€ν…(50Hz, dt=0.02s)별 μƒνƒœ-행동(State-Action) 쌍으둜 κ΅¬μ„±λ©λ‹ˆλ‹€.
48
+
49
+ | 파일λͺ… | ν˜•μ‹ | μ„€λͺ… |
50
+ | :--- | :---: | :--- |
51
+ | `data_part_01.csv` ~ `data_part_10.csv` | CSV | 총 250,000 ν”„λ ˆμž„ μ›μ‹œ μ œμ–΄ 둜그 (14κ΄€μ ˆκ°, 토크, GRF $F_x/F_z$, Slip 속도) |
52
+ | `v1.0_baseline_dataset_metadata.json` | JSON | 데이터셋 곡식 메타데이터 및 벀치마크 μ§€ν‘œ μŠ€ν‚€λ§ˆ |
53
+
54
+ ---
55
+
56
+ ## πŸ—ΊοΈ Roadmap & Future Updates (ν–₯ν›„ μ—…λ°μ΄νŠΈ κ³„νš)
57
+
58
+ ν˜„μž¬ **V1.0**은 κ°€μž₯ 근간이 λ˜λŠ” **'κΈ°μ€€ 보행(Baseline Locomotion)'**을 μ œκ³΅ν•©λ‹ˆλ‹€. λ‹€μŒ μ—…λ°μ΄νŠΈλ₯Ό 톡해 μƒμš© λ‘œλ΄‡ μ œμ–΄μ˜ 전체 μƒνƒœκ³„λ₯Ό μ œκ³΅ν•  μ˜ˆμ •μž…λ‹ˆλ‹€.
59
+
60
+ * **[Current] V1.0**: Zero-Slip 평면 보행 κ΅°μ§‘ 데이터.
61
+ * **[Upcoming] V2.0**: μ™„μ „ 자율 μƒνƒœ 기계(No-Button Autopilot FSM) 연동. μ™ΈλΆ€ μ™Έλž€ 감지 μ‹œ μžλ™ 착석(Auto-Sit), 기립(Auto-Stand) 및 λ‚™ν•˜ μžκ°€ 볡ꡬ(Auto-Recovery) μ œμ–΄ 데이터 νƒ‘μž¬ μ˜ˆμ •.
62
+ * **[Upcoming] V3.0**: 레이캐슀트(Raycast) λΉ„μ „ μ„Όμ„œ 톡합. μ „λ°© μž₯μ• λ¬Ό λŠ₯동 인지, μ‘°ν–₯ νšŒν”Ό(Active Avoidance) 및 동적 κ· ν˜• 회볡 데이터 μΆ”κ°€.
data_part_01.csv ADDED
The diff for this file is too large to render. See raw diff
 
data_part_02.csv ADDED
The diff for this file is too large to render. See raw diff
 
data_part_03.csv ADDED
The diff for this file is too large to render. See raw diff
 
data_part_04.csv ADDED
The diff for this file is too large to render. See raw diff
 
data_part_05.csv ADDED
The diff for this file is too large to render. See raw diff
 
data_part_06.csv ADDED
The diff for this file is too large to render. See raw diff
 
data_part_07.csv ADDED
The diff for this file is too large to render. See raw diff
 
data_part_08.csv ADDED
The diff for this file is too large to render. See raw diff
 
data_part_09.csv ADDED
The diff for this file is too large to render. See raw diff
 
data_part_10.csv ADDED
The diff for this file is too large to render. See raw diff
 
v1.0_baseline_dataset_metadata.json ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset_name": "MicroDuck-V1.0-FlatGround-Baseline",
3
+ "version": "1.0.0",
4
+ "release_date": "2026-09-03",
5
+ "license": "Commercial / Proprietary",
6
+ "hardware_platform": "Antigravity MicroDuck 14-DOF Bipedal Robot",
7
+ "dof_configuration": {
8
+ "total_dof": 14,
9
+ "legs": "2 x 5-DOF (HipYaw, HipRoll, HipPitch, KneePitch, AnklePitch)",
10
+ "head_neck": "3-DOF (NeckPitch, HeadPitch, HeadYaw)",
11
+ "tail": "1-DOF"
12
+ },
13
+ "telemetry_specs": {
14
+ "control_frequency_hz": 50.0,
15
+ "total_episodes": 500,
16
+ "total_agents": 10,
17
+ "total_frames": 250000,
18
+ "ground_surface": "Flat Rigid Floor (MuJoCo Ground Plane)"
19
+ },
20
+ "performance_benchmarks": {
21
+ "mean_forward_velocity_m_s": 0.652,
22
+ "mean_stance_slip_velocity_m_s": 0.0003194,
23
+ "max_stance_slip_velocity_m_s": 0.002121,
24
+ "ground_clearance_swing_mm": 22.0,
25
+ "mean_grf_vertical_n": 9.667,
26
+ "mean_grf_propulsion_fx_n": 1.35,
27
+ "collision_rate_percent": 0.0,
28
+ "stagnation_marching_rate_percent": 0.0
29
+ },
30
+ "files": {
31
+ "raw_dataset_csv": "v1.0_baseline_10ducks_500ep.csv",
32
+ "model_weights_onnx": "v1.0_flat_ground_stable.onnx",
33
+ "metadata_json": "v1.0_baseline_dataset_metadata.json"
34
+ }
35
+ }