Policy Inference

1. Policy Inference

Run inference using a trained policy model and automatically record the results:

/isaac-sim/python.sh src/lerobot/scripts/lerobot_record.py \
    --robot.type=walker_s2_sim \
    --robot.headless=false \
    --task=Part_Sorting \
    --policy.path=your_checkpoint_dir \
    --dataset.repo_id=your_org/eval_Part_Sorting \
    --dataset.single_task="Part Sorting" \
    --dataset.num_episodes=2 \
    --dataset.push_to_hub=false \
    --dataset.episode_time_s=100000000 \
    --dataset.num_image_writer_processes=4 \
    --dataset.root=${workspaceFolder}/datasets/Part_Sorting/your_policy_type \
    --dataset.video=true \
    --play_sounds=false

Parameter	Description	Default / Notes
--robot.type	Robot type	walker_s2_sim
--robot.headless	Whether to run in headless mode	false
--task	Task name (Part_Sorting, Conveyor_Sorting, Foam_Inlaying, or Packing_Box)	Foam_Inlaying
--policy.path	Path to the policy model checkpoint, either a local path or a Hugging Face Hub repository	Required
--dataset.repo_id	Dataset identifier. In inference mode, it should follow the xx/eval_xx format	Required
--dataset.single_task	Task description	Required unless automatically set based on --task
--dataset.num_episodes	Number of inference episodes	50
--dataset.root	Directory where the inference results are saved	Required
--dataset.video	Whether to record videos	true
--dataset.fps	Recording frame rate	30
--dataset.episode_time_s	Duration of each episode, in seconds	60
--dataset.push_to_hub	Whether to upload the dataset to the Hugging Face Hub	false
--play_sounds	Whether to play audio prompts	true

Notes:

1. Policy inference is performed using the lerobot_record.py script. Specify the policy model checkpoint using the --policy.path parameter. The script automatically loads the model, runs inference in the simulation environment, and records the results.
2. In inference mode, the --dataset.repo_id parameter should follow the xx/eval_xx format, such as your_org/eval_Part_Sorting, to distinguish evaluation datasets from datasets created in data collection mode.
3. Replace ${workspaceFolder} with the actual path to your workspace.