Views
No views yet
full_20260529_0220, 45-task eval_v1_100 set). Inference-only
(lerobot pretrained_model/ layout: weights + config + pre/post-processors).| Base model | trained from scratch (no pretrained init) |
| Dataset | zhouqh/hrbench → genesis_hr_bench_lerobot_qpos_target_abs_20260526_1651 |
| Action space | qpos_abs (8-D Franka joint absolute + gripper) |
| Step | 35,000 |
| Wandb | act_genesis_hr_bench/cwc1tdbe |
| Uploaded | 2026-08-05 06:59:28 UTC |
Note. Configured for 100k steps but the final saved checkpoint is step 35000 (checkpoints/last -> 035000). wandb run was resumed once (two local session dirs share id cwc1tdbe).
1from lerobot.policies.factory import make_policy_from_pretrained
2
3policy = make_policy_from_pretrained("zimplex/act-genesis-hr-bench-qpos-abs-joint-step35000", device="cuda")
4policy.eval()
5# Observation keys: observation.images.base_0_rgb / left_wrist_0_rgb /
6# right_wrist_0_rgb (as available), observation.state. Action space: qpos_abs.
7action_chunk = policy.predict_action(batch_observation)Pair the checkpoint with the matchingqpos_absdataset variant — the normalizers andaction_dimare baked in.