rsl_rl

rsl_rl is a fast, GPU-native reinforcement-learning training library (PPO and friends) from Marco Hutter's Robotic Systems Lab. It is the default RL trainer of NVIDIA's Isaac Lab ecosystem, giving it massive downstream adoption across sim-to-real legged- and manipulation-robot research (~2.8K★). No single canonical paper — a GitHub-verified framework underpinning the group's ANYmal locomotion work.