Zero-HLM: Generalizable Humanoid Loco-Manipulation without Real Robot Data
Paper • 2610.35075 • Published
task stringclasses 3
values | episodes int64 800 800 | frames int64 648k 963k | archive stringclasses 3
values | archive_bytes int64 2.59B 3.02B | sha256 stringclasses 3
values | url stringclasses 3
values |
|---|---|---|---|---|---|---|
task1 | 800 | 648,370 | zero_hlm_bench_task1_800.tar | 2,791,987,200 | a4e122b89089b7d7eff38ee7de1912ac3a743decd1333665be83566124a00072 | https://huggingface.co/datasets/YDY0427/Zero-HLM-Bench/resolve/main/zero_hlm_bench_task1_800.tar |
task2 | 800 | 750,245 | zero_hlm_bench_task2_800.tar | 3,024,332,800 | f1eee1882f800925d0f3469737b024e4c142c47a13f312d4433c79f8d3ee64d9 | https://huggingface.co/datasets/YDY0427/Zero-HLM-Bench/resolve/main/zero_hlm_bench_task2_800.tar |
task3 | 800 | 963,304 | zero_hlm_bench_task3_800.tar | 2,589,358,080 | 36ebef9fdf109493e4e6e3fd625c5f09a3aa1f264da0956080684f905aa36494 | https://huggingface.co/datasets/YDY0427/Zero-HLM-Bench/resolve/main/zero_hlm_bench_task3_800.tar |
Zero-HLM-Bench
Training demonstrations for humanoid loco-manipulation in three simulated tasks. Each task contains 800 episodes in LeRobot v2.1 format.
| Task | Description | Episodes | Frames | Download |
|---|---|---|---|---|
| Task 1 | Walk, pick, and place | 800 | 648,370 | Dataset |
| Task 2 | Place a hat on a cabinet | 800 | 750,245 | Dataset |
| Task 3 | Carry a basket through a door | 800 | 963,304 | Dataset |
Data Format
Episodes contain synchronized 50 Hz RGB observations, 34-dimensional robot states, and 66-dimensional actions consisting of 64 SONIC latent values and two hand commands. RGB videos are 640 × 480 H.264 files without audio. Observations combine a Gaussian-splatting background with a mesh robot and task objects.
Each archive extracts into task1/, task2/, or task3/, containing data/
(Parquet episodes), videos/ (RGB observations), and meta/ (feature schema,
episode records, task instructions, and statistics). Episode indices run from
0 to 799 in each task. Per-file hashes are provided in each archive's files.json.
Download
hf download YDY0427/Zero-HLM-Bench --repo-type dataset --local-dir training_data
cd training_data
sha256sum --check SHA256SUMS
tar -xf zero_hlm_bench_task1_800.tar
tar -xf zero_hlm_bench_task2_800.tar
tar -xf zero_hlm_bench_task3_800.tar
- Downloads last month
- 102