kms0429/xle_pilot_onearm_dice_front_21ep
LeKiwi โข Updated โข 21 episodes โข 123
How to use mongster7/smolvla_dice_50k with LeRobot:
# See https://github.com/huggingface/lerobot?tab=readme-ov-file#installation for more details git clone https://github.com/huggingface/lerobot.git cd lerobot pip install -e .[smolvla]
# Launch finetuning on your dataset python lerobot/scripts/train.py \ --policy.path=mongster7/smolvla_dice_50k \ --dataset.repo_id=lerobot/svla_so101_pickplace \ --batch_size=64 \ --steps=20000 \ --output_dir=outputs/train/my_smolvla \ --job_name=my_smolvla_training \ --policy.device=cuda \ --wandb.enable=true
# Run the policy using the record function
python -m lerobot.record \
--robot.type=so101_follower \
--robot.port=/dev/ttyACM0 \ # <- Use your port
--robot.id=my_blue_follower_arm \ # <- Use your robot id
--robot.cameras="{ front: {type: opencv, index_or_path: 8, width: 640, height: 480, fps: 30}}" \ # <- Use your cameras
--dataset.single_task="Grasp a lego block and put it in the bin." \ # <- Use the same task description you used in your dataset recording
--dataset.repo_id=HF_USER/dataset_name \ # <- This will be the dataset name on HF Hub
--dataset.episode_time_s=50 \
--dataset.num_episodes=10 \
--policy.path=mongster7/smolvla_dice_50klerobot/smolvla_base๋ฅผ XLeRobot(LeKiwi) ๋ชจ๋ฐ์ผ ๋งค๋ํฐ๋ ์ดํฐ ์์ฐ ๋ฐ์ดํฐ๋ก ๋ฏธ์ธ์กฐ์ ํ ์ ์ฑ
์
๋๋ค.
Task: Put the red dice in the basket
| ํญ๋ชฉ | ๊ฐ |
|---|---|
| Base checkpoint | lerobot/smolvla_base |
| Dataset | kms0429/xle_pilot_onearm_dice_front_21ep |
| Episodes / frames | 21 / 26,178 (30 fps) |
| Robot | lekiwi_client โ 9-D stateยทaction (arm 6-DoF + base 3-DoF) |
| Camera | 1ร front, 640ร480 |
| Steps | 50,000 (โ15.3 epochs) |
| Batch size | 8 |
| Trainable params | 99.9M / 450M (22%) |
| Frozen | vision encoder, VLM backbone (train_expert_only=True) |
| Precision | AMP |
| Hardware | 1ร RTX 5070 Laptop (8GB) |
| Wall-clock | 2h 45m |
| step | loss |
|---|---|
| 100 | 1.973 |
| 5,000 | 0.138 |
| 15,000 | 0.067 |
| 25,000 | 0.044 |
| 50,000 | 0.042 |
Loss๋ ์ฝ 25,000 step์์ ์๋ ดํ์ต๋๋ค. ๊ฐ์ ๊ท๋ชจ๋ก ์ฌํํ์ค ๊ฒฝ์ฐ 25,000 step์ผ๋ก๋ ์ถฉ๋ถํฉ๋๋ค.
lerobot-record \
--robot.type=lekiwi_client \
--robot.remote_ip=<robot-ip> \
--policy.path=mongster7/smolvla_dice_50k \
--dataset.single_task="Put the red dice in the basket"
์นด๋ฉ๋ผ ํค๊ฐ observation.images.front์ธ ๋ฐ์ดํฐ์
์ผ๋ก ํ์ตํ์ผ๋ฏ๋ก, ์ถ๋ก ์์๋ ๋์ผํ rename์ด ํ์ํฉ๋๋ค.
--rename_map='{"observation.images.front": "observation.images.camera1"}'
smolvla_base์ ์ฌ์ ํ์ต ๋ถํฌ(6-DoF ๊ณ ์ ํ)์ ํฌํจ๋์ง ์์, 21 ์ํผ์๋๋ง์ผ๋ก ํ์ต๋์์ต๋๋ค. ํ ๋์ ๋๋น ์ด๋ ์ฑ๋ฅ์ด ๋ฎ์ ์ ์์ต๋๋ค.Base model
HuggingFaceTB/SmolLM2-360M