RLRoverLAB: An Advanced Reinforcement Learning Suite for Planetary Rover Simulation and Training

This paper presents RLRoverLAB, an open-source reinforcement learning suite tailored for simulating planetary rovers on extraterrestrial bodies. The suite features a set of space related assets and tasks implemented using the Nvidia ORBIT framework, backed by a robust physics and graphics engine in Nvidia Omniverse. The suite aims to bridge the gap between space robotics and reinforcement learning, offering researchers a suite of assets, predefined tasks, and a versatile platform for developing, testing and benchmarking novel reinforcement learning algorithms. Consequently, the suite is designed to be flexible and modular, allowing for easy integration of new assets and tasks. Through exemplary use cases and tasks, we demonstrate RLRoverLAB’s potential to advance RL applications in space exploration. Videos, documentation, and code available is at https://github.com/abmoRobotics/isaac_rover_orbit