Simulation and Reinforcement Learning

Two simulation / reinforcement-learning pipelines for the AgiBot X2 — pick one based on your use case

This chapter brings together two independent pipelines related to simulation and reinforcement learning on the AgiBot X2. They target different use cases and use different code repositories and ways of running; they do not depend on each other — pick the one you need. Do not drive the same simulation or the same robot with more than one pipeline at the same time.

Pipeline	Use case	Language / entry point	Code location
X2 RL Deployment	Deploy an existing RL model as an example (the bundled model is a dance motion)	C++ `motion_control_node.cc`	`extra/x2_rl_deploy/` (git submodule)
X2 Sim + MC Deployment	Drive the MuJoCo simulation with the official motion-control module `mc`, running the same stand / walk / get-up / animation skills as on the real robot	SDK examples (`py_examples` / `examples`)	`mc` + `sim_mujoco`

How to choose

You only want to see how to wire an onnx model into the SDK and prefer C++ → X2 RL Deployment
You do not do reinforcement learning and just want to validate your own high-level logic in simulation on an x86 host (competition simulation, algorithm integration) → X2 Sim + MC Deployment

Note

The examples of both pipelines have been validated only in simulation; the motor KP/KD values are simulation recipes — you must tune the parameters yourself before using them on a real robot.