X2 RL Deployment

Note: The example program has only been validated in a simulation environment. To use it to control a real AgiBot X2 robot, parameter re-tuning is required.

• Provides a minimal example for deploying a user-trained RL model.

• The RL model used in this example performs a dance motion, and reference code for model control and a simple inference engine is provided.

• After completing this example, you can replace the onnx model file and adapt your policy in the RlDefault function in motion_control_node.cc to deploy a custom model.

Note

When to use this chapter: wiring an existing onnx model into the SDK, implemented in C++ (code lives in extra/x2_rl_deploy/). The bundled model is a dance motion; training is not covered here.

• 1 Usage Instructions
  • 1.1 Prerequisites
  • 1.2 Quick Start
    • Launch the motion-control example node
    • Launch the simulation node
    • Launch the gamepad node
  • 1.3 Run the RL model example
    • Button Mapping
  • 1.4 Directory Structure
  • 1.5 Model Deployment
  • 1.6 Communication Protocol
    • Communication Topology
    • Joint State Subscription Topics
    • IMU State Subscription Topics
    • Control Command Publication Topics