skrl

A modular Python reinforcement learning library that plugs directly into Isaac Lab.

skrl is an open-source Python reinforcement learning library started by Antonio Serrano-Muñoz, supporting both PyTorch and JAX backends. It splits agents, memory (experience storage), models, and trainers into separate, swappable modules, and includes common algorithms such as PPO (Proximal Policy Optimization), SAC, and TD3 out of the box. Its defining feature is native support for Gymnasium as well as NVIDIA's GPU-parallel environments, Isaac Gym and Isaac Lab; it's one of the few reinforcement learning libraries officially supported by Isaac Lab, alongside rsl_rl, rl_games, and Stable-Baselines3.

ExampleWhen training quadruped locomotion in Isaac Lab, swap the training script's library argument from rsl_rl to skrl, and run the same task with its PPO implementation instead.

Related

NVIDIA Isaac Lab · rsl_rl · rl_games · Stable-Baselines3 · Proximal Policy Optimization · Gymnasium

Sources

skrl documentation
Toni-SM/skrl (GitHub)