Views
No views yet
smolvla), ~450M parameters, safetensorscamera1, camera2, camera3)n_action_steps=50)| Setting | Value |
|---|---|
| Dataset | Qiu-Xinchuan/so101_Multi-object-packing |
| Steps | 100,000 |
| Batch size | 8 |
| Learning rate | 1e-4 |
| Seed | 1000 |
1lerobot-train \
2 --policy.type=smolvla \
3 --dataset.repo_id=Qiu-Xinchuan/so101_Multi-object-packing \
4 --steps=100000 \
5 --batch_size=8 \
6 --policy.optimizer_lr=1e-4 \
7 --seed=1000 \
8 --job_name=smolvla_packing1lerobot-eval \
2 --policy.path=Porameht/smolvla_so101_packing_reproduce \
3 --env.type=... # or run on a real SO101 via lerobot-record / robot control scripts1from lerobot.policies.factory import make_policy
2from lerobot.configs.policies import PreTrainedConfig
3
4config = PreTrainedConfig.from_pretrained("Porameht/smolvla_so101_packing_reproduce")
5policy = make_policy(config, ds_meta=...) # dataset meta from Qiu-Xinchuan/so101_Multi-object-packing1@article{shukor2025smolvla,
2 title={SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics},
3 author={Shukor, Mustafa and others},
4 journal={arXiv preprint arXiv:2506.01844},
5 year={2025}
6}