Custom package based on lerobot. The package is used to control the arm So100, but we implement our own pipeline to control the robot, create the dataset and to train a policy.
The following dependencies are required to use this package:
- Docker
- NVIDIA Container Toolkit (if you are using an NVIDIA GPU)
- LeRobot arm
The easiest way to get started is to use the provided Docker image. You can find the Dockerfile in the docker folder. To build the image, run the following command:
sh build.shTo run the development container, use the following command:
sh run.shWe prepare a script to teleoeprate the robot. You can use the following command to run the script:
python lerobot_control.pyIt's important to note that the teleoperation it used the zero position as the one first detected. So the idea is tu put both arm in the same position and then run the script. The script will detect the zero position and use it as the reference to send the delta movement for each joint.
We provide a script to record the dataset. The script will record the dataset in the dataset folder with the task_name as a subfolder. The dataset will be recorded in the following format:
python lerobot_control.py --dataset_path="Simple cross motion"Where:
--dataset_task="Simple cross motion": This is the name of the task that will be used to create the subfolder in the dataset folder.
We provide a script to replay a specific episode of the dataset. The script will replay the dataset in the dataset folder with the task_name as a subfolder. The dataset will be replayed in the following format:
python lerobot_control.py --dataset_path="Simple cross motion" --replay_episode=00000Where:
--dataset_task="Simple cross motion": This is the name of the task to be replayed.--replay_episode=00000: This is the name of the episode to be replayed. The episode name is the same as the one used to record the dataset.
TODO
TODO