rex-gym

Why we included this project

For anyone learning to combine reinforcement learning with legged robots, Rex offers a practical entry point. It provides OpenAI Gym environments for a 3D-printed SpotMicro quadruped, simulated in PyBullet, so you can train control policies before touching physical hardware. The repo includes the robot's URDF model, a PPO implementation, and a command-line tool that handles single or batch training and lets you play back a pretrained policy. You can compare two control approaches, an open-loop gait and a Bezier inverse-kinematics controller, which helps clarify what actually shapes locomotion. The project is designed for sim-to-real transfer on low-cost quadruped platforms, and the codebase is small enough to serve as a concrete reference.