XLeRobot Documentation

XLeRobot runs under the lerobot framework. Users can choose from several control methods depending on their hardware and task complexity:

• Joint Control: Used for basic motor testing.
• End-Effector Control: Used for teleoperation.
• Advanced Control: Includes dual-arm control or vision-based control.
• Input Methods: Full system teleoperation can be achieved using a keyboard, Xbox controller, or Switch Joycon.

All example scripts are located in the software/examples/ directory and can be run after proper setup and calibration. Note that some examples require additional calibration for optimal performance.

XLeRobot supports various simulation environments for scene manipulation and robotics tasks. The following environments can be used:

• ReplicaCAD_SceneManipulation-v1 (the default environment)
• AI2THOR scenes
• Robocasa Kitchen counter scenes
• OpenCabinetDrawer-v1

XLeRobot requires a power station or power bank capable of supporting at least two 60W+ USB-C fast charging ports to power the 12V arms, base, and Raspberry Pi.

Recommended Setup:

• Anker SOLIX C300 Power Station: Provides 288Wh capacity and 300W max output. It can power the entire system (including dual arms and Raspberry Pi, which draw ~180W max) using three USB-C cables, which simplifies wiring.

Alternative Compatible Options (Not Tested):

• UGREEN 300W 48000mAh Power Bank
• Anker Laptop Power Bank (25K, 165W)
• AOHI 240W Laptop Power Bank 27600mAh

To run the ZMQ host on the actual robot hardware, use the xlerobot_2wheels_host module. Ensure PYTHONPATH is set to include the src directory.

PYTHONPATH=src python -m lerobot.robots.xlerobot_2wheels.xlerobot_2wheels_host --robot.id=my_xlerobot_2wheels

XLeRobot 0.1.5 provides a simulation environment (Playground) that allows you to reproduce demos in approximately 10 minutes. It includes all necessary URDF files and control scripts.

To get started with simulation, refer to the Step-by-step Installation Guide in the simulation/sim_guide_CN.md file and the technical blog in simulation/sim_CN.md.

Before running SmolVLA or ACT, ensure the following environment and hardware configurations are met:

1. Environment: Activate the conda environment using conda activate lerobot.
2. Cameras: Verify that three cameras are connected and their indices are correctly identified (typically 0, 1, and 2).
3. Follower Ports: Confirm the follower ports are correctly mapped (e.g., /dev/ttyACM0, /dev/ttyACM1).
4. Leader Ports: Confirm the leader ports are correctly mapped (e.g., /dev/ttyACM2, /dev/ttyACM3).
5. Dataset: Ensure the dataset repository ID and the task description are properly set.
6. Training: Verify that training is running on the intended device (cuda vs cpu).
7. Inference: Ensure inference is pointed to the correct path of the trained policy.

Developer assembly kits (excluding battery and IKEA cart) are available for purchase through Wowrobo.

• Worldwide: $579 via Wowrobo Shop
• China: 3699￥ via Taobao

To set up the XLeRobot simulation environment, follow these steps:

1. Prerequisites: Ensure you are running Ubuntu and have Miniconda installed.
2. Create Conda Environment: Create a Python 3.11 environment named lerobot.
3. Install ManiSkill: Install the mani-skill package via pip and download the ReplicaCAD scene dataset.
4. Install Additional Dependencies: Install pygame for keyboard input/control panels and rerun-sdk for camera data visualization.

# Create and activate environment
conda create -y -n lerobot python=3.11
conda activate lerobot

# Install ManiSkill and dataset
pip install mani-skill
python -m mani_skill.utils.download_asset "ReplicaCAD"

# Install additional dependencies
pip install pygame
pip install rerun-sdk