Watch the demo:
We collected three datasets:
And we merged them the last two to a single LeRobot dataset dopaul/merged_grabbing_dataset_v7
We trained them using the script below. If you don't pass a wandb api key then set wandb.enable to false.
cd lerobot
WANDB_API_KEY=[YOUR_WANDB_API_KEY] python lerobot/scripts/train.py \
--policy.type=act \
--dataset.repo_id=dopaul/merged_grabbing_dataset_v7 \
--wandb.enable=true \
--steps=100000 \
--output_dir=lerobot/outputs/train/zrh_grasping_v7
# Upload model to HF
python lerobot/scripts/push_pretrained.py \
--pretrained_path=lerobot/outputs/train/zrh_grasping_v7/checkpoints/last/pretrained_model \
--repo_id=dopaul/zrh_grasping_v7
- Copy
.env.exampleand rename it to.env. Create an OpenAI key and replace the placeholder. - Install packages by running
poetry install - Run
main.py