python lerobot/scripts/train.py \
  --dataset.repo_id=<user_or_org>/<dataset> \
  --policy.type=act \
  --output_dir=outputs/train/<desired_policy_repo_id> \
  --job_name=lerobot_training \
  --policy.device=cuda \
  --policy.repo_id=<user_or_org>/<desired_policy_repo_id> \
  --wandb.enable=true

Writes checkpoints to outputs/train/<desired_policy_repo_id>/checkpoints/.

Evaluate the policy

bash

python -m lerobot.record \
  --robot.type=so100_follower \
  --dataset.repo_id=<user_or_org>/eval_<dataset> \
  --policy.path=<user_or_org>/<desired_policy_repo_id> \
  --episodes=10

Prefix the dataset repo with eval_ and supply --policy.path pointing to a local or hub checkpoint.

hoon-shin/eval_act_so100_test

Model Card for eval_act_so100_test

How to Get Started with the Model

Train from scratch

Evaluate the policy

hoon-shin/
eval_act_so100_test