SKRL is a modular reinforcement learning library for researchers and practitioners, offering a flexible and transparent implementation of RL algorithms. It supports PyTorch and JAX, integrates with popular environment interfaces like Gymnasium and Brax, and provides specialized support for NVIDIA's Isaac Gym, Omniverse Isaac Gym, and Isaac Lab, enabling efficient, parallelized training.

How It Works

SKRL is built on a modular architecture, allowing users to easily swap components and experiment with different RL algorithms. It leverages PyTorch and JAX for efficient tensor operations and automatic differentiation. A key advantage is its ability to manage and train agents across multiple, potentially resource-sharing, subsets of environments simultaneously, significantly accelerating the training process for complex robotics and simulation tasks.

Quick Start & Requirements

• Install via pip: pip install skrl
• Requires Python 3.8+
• For NVIDIA Isaac Gym/Lab integration, CUDA Toolkit and NVIDIA drivers are necessary.
• Documentation: https://skrl.readthedocs.io

Highlighted Details

• Supports PyTorch and JAX backends.
• Integrates with Gymnasium, PettingZoo, Brax, and NVIDIA Isaac Gym/Lab.
• Enables simultaneous training across environment subsets (scopes).
• Provides a modular and transparent implementation of various RL algorithms.

Maintenance & Community

The project is under active development. Citation information is available via a JMLR publication.

Licensing & Compatibility

The library is released under the MIT License, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

The project is under continuous development, and users are advised to use the latest versions for updates. Specific NVIDIA Isaac Gym/Lab features may require compatible hardware and software versions.