中文版本：

构建数据集训练模型并推理

Back to Table of Contents:

Getting Started

Collect a Dataset

On your AhaRobot host machine, run the following to collect a dataset:

# rm -rf /home/rosdev/.cache/huggingface/lerobot/lookas/astra_grab_floor_toys_extended # if you want to start over
# rm -rf /home/rosdev/.cache/huggingface/lerobot/lookas/astra_grab_floor_toys_extended/images # if collection was interrupted
pushd non_ros_src/lerobot
python lerobot/scripts/control_robot.py \
  --robot.type=astra_joint \
  --control.type=record \
  --control.single_task="Grab the toys on the floor and put them on the tabletop." \
  --control.fps=30 \
  --control.repo_id=lookas/astra_grab_floor_toys_extended \
  --control.tags='["astra"]' \
  --control.warmup_time_s=0 \
  --control.episode_time_s=-1 \
  --control.reset_time_s=-1 \
  --control.num_episodes=50 \
  --control.push_to_hub=true \
  --control.resume=false \
  --control.local_files_only=true \
  --control.run_compute_stats=true

Notes:

• –control.num_episodes=50 sets the number of demos to collect.
• –control.push_to_hub=false prevents uploading data to Hugging Face.
• –control.resume=true appends to an existing dataset.
• –control.run_compute_stats=false skips dataset statistics to speed up collection.

Inspect Dataset Contents

After collection, you can use rerun to visualize the dataset:

pushd non_ros_src/lerobot
python lerobot/scripts/visualize_dataset.py --repo-id lookas/astra_grab_floor_toys_extended --episode-index 51 --local-files-only 1

⚠️ If you see 3D/GPU errors when running rerun inside Docker, launch rerun outside Docker.

Train Your Model

Train on your collected dataset: