autorl-grab-act-vibepi-iter10-aug
act policy trained via Vibe Data Studio on 195 teleop episodes.
At a glance
| Architecture | act (from scratch) |
| Dataset | grab-iter10-iter5basis-augmented-v1-flat @ 4ba83ba1 |
| Scale | 195 episodes · 16,070 frames @ 20 fps |
| Steps / batch / seed | 60,000 / 8 / None |
| lerobot | 30da8e68 |
| Provider / GPU | runpod / NVIDIA A40 |
| Duration | 3h 24m |
| Final train loss | 0.069 |
| W&B | vds45-627aef3d |
Tasks
Language-conditioned — pass one of these instructions at inference:
| Instruction | Episodes |
|---|---|
| grab the cucumber close to one of the cucumber's end | 182 |
| 000004-cucumber-grab | 1 |
| 000005-cucumber-grab | 1 |
| 000007-cucumber-grab | 1 |
| 000010-cucumber-grab | 1 |
| 000011-cucumber-grab | 1 |
| 000014-cucumber-grab | 1 |
| 000015-cucumber-grab | 1 |
| 000016-cucumber-grab | 1 |
| 000018-cucumber-grab | 1 |
| 000019-cucumber-grab | 1 |
| 000020-cucumber-grab | 1 |
| 000021-cucumber-grab | 1 |
| 000022-cucumber-grab | 1 |
Rig
- Action/state (7-dim): shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper, tilt
- Cameras @ 20 fps: base 640×480, top 640×480, wrist 640×480
Data provenance
Materialized collection grab-iter10-iter5basis-augmented@v1 (frozen 2026-07-26):
cucumber-grab-DAgger-iter4-v2-trim@34b0f3ab(182 eps)autoresearch-000004-cucumber-grab@e2e071d9(1 eps)autoresearch-000005-cucumber-grab@22dd4341(1 eps)autoresearch-000007-cucumber-grab@72f846fd(1 eps)autoresearch-000010-cucumber-grab@c9dd850b(1 eps)autoresearch-000011-cucumber-grab@d91e0655(1 eps)autoresearch-000014-cucumber-grab@718289c1(1 eps)autoresearch-000015-cucumber-grab@8ce7acae(1 eps)autoresearch-000016-cucumber-grab@3582067c(1 eps)autoresearch-000018-cucumber-grab@fd7912d9(1 eps)autoresearch-000019-cucumber-grab@3de26824(1 eps)autoresearch-000020-cucumber-grab@975ab111(1 eps)autoresearch-000021-cucumber-grab@3b4af4e4(1 eps)autoresearch-000022-cucumber-grab@8e0479ac(1 eps)
lerobot-train --policy.type=act --policy.repo_id=VibeCuisine/autorl-grab-act-vibepi-iter10-aug --dataset.repo_id=VibeCuisine/grab-iter10-iter5basis-augmented-v1-flat --dataset.revision=4ba83ba14f9aa6493e93ed4ed4f641752e2007f8 --batch_size=8 --steps=60000 --output_dir=outputs/train/autorl-grab-act-vibepi-iter10-aug --job_name=grab-iter10-iter5basis-augmented --policy.device=cuda --wandb.enable=true --wandb.disable_artifact=true --wandb.project=lerobot --wandb.run_id=vds45-627aef3d --wandb.entity=jeremyhx-freelance --dataset.image_transforms.enable=true
--num_workers is appended host-side (clamped to the host's allocation).
Card generated by Vibe Data Studio on run completion.