Views
No views yet
lerobot/smolvla_base
on the CoRL2026-CSI/UR7e_CaP_PickandPlace_100epi_10fps
dataset for a UR7e single-arm pick-and-place task.| Field | Value |
|---|---|
| Base model | lerobot/smolvla_base |
| Dataset | CoRL2026-CSI/UR7e_CaP_PickandPlace_100epi_10fps (100 eps, 35,878 frames, 10 fps) |
| Robot | UR7e single-arm, 7-DoF (6 joints + gripper) |
| Cameras | realsense_topview, realsense_wrist (renamed → camera1/camera2) |
| Steps | 7,100 (≈ 50 epoch · 35878 × 50 / 256) |
| Batch | 128 × 2 GPU = 256 per-step samples |
| Optimizer | AdamW (lr 1e-4, betas (0.9, 0.95), wd 1e-10), cosine decay w/ warmup 1000 |
| Chunk / Action steps | 50 / 50 |
| Image augmentation | brightness, contrast, saturation, hue, sharpness, affine (max 3, random order) |
| Hardware | 2× NVIDIA RTX PRO 6000 Blackwell |
action/observation.state dim 은 7 이며, SmolVLA 의 max_action_dim=32 으로 자동 zero-pad 됩니다.1import torch
2from lerobot.policies.smolvla.modeling_smolvla import SmolVLAPolicy
3
4policy = SmolVLAPolicy.from_pretrained("CoRL2026-CSI/smolVLA-UR7e-PickandPlace-50epoch")
5policy.to("cuda").eval()
6
7# observation 의 카메라 키는 학습 시 사용한 이름(`observation.images.camera1`,
8# `observation.images.camera2`) 과 동일해야 합니다.
9with torch.inference_mode():
10 action = policy.select_action(observation)1lerobot-train \
2 --policy.path=CoRL2026-CSI/smolVLA-UR7e-PickandPlace-50epoch \
3 --dataset.repo_id=CoRL2026-CSI/UR7e_CaP_PickandPlace_100epi_10fps \
4 --output_dir=outputs/train/smolvla_ur7e_pickandplace_ft \
5 --job_name=smolvla_ur7e_pickandplace_ft \
6 --batch_size=128 --steps=2000 \
7 --policy.device=cuda --wandb.enable=truescripts/cap/smolvla_cap_ur7e_pickandplace.sh 이며,
정확한 hyperparameter 는 이 리포의 train_config.json 으로도 재구성 가능합니다.lerobot/smolvla_base