Toni-SM/skrl

What it solves

skrl provides a modular and transparent way to implement and use Reinforcement Learning (RL) algorithms. It simplifies the process of training agents across various environment interfaces, allowing for flexible configuration and simultaneous training across subsets of environments (scopes) in a single run.

How it works

The library is implemented in Python and supports multiple backends including PyTorch, JAX, and NVIDIA Warp. It integrates with a wide range of environment interfaces such as OpenAI Gym, Farama Gymnasium, PettingZoo, ManiSkill, NVIDIA Isaac Lab, and MuJoCo Playground.

Who it’s for

It is designed for researchers and developers who need a modular, readable, and simple RL library that can handle diverse environments and supports multiple deep learning frameworks.

Highlights

• Multi-backend support: Works with PyTorch, JAX, and NVIDIA Warp.
• Hugging Face integration: Models are available on Hugging Face.
• Broad environment compatibility: Supports Gymnasium, PettingZoo, ManiSkill, and specialized simulators like Isaac Lab and MuJoCo Playground.
• Flexible training: Enables simultaneous training of agents by scopes (subsets of environments) within the same run.