rllab is a framework for developing and evaluating reinforcement learning algorithms, primarily for continuous control tasks. It was designed for researchers and practitioners in RL, offering a structured approach to algorithm implementation and experimentation, with tools for distributed execution and visualization.

How It Works

rllab provides a modular structure for RL algorithms, abstracting common components like policies, value functions, and environments. It leverages Theano as its primary backend for automatic differentiation and computation, with experimental TensorFlow support available in a separate module. This design facilitates the implementation and comparison of various RL algorithms.

Quick Start & Requirements

• Install via pip: pip install rllab
• Requires Python 3.5+. An older Python 2 snapshot is available on the py2 branch.
• Official documentation: https://rllab.readthedocs.org/en/latest/

Highlighted Details

• Implements algorithms: REINFORCE, Truncated Natural Policy Gradient, Reward-Weighted Regression, Relative Entropy Policy Search, Trust Region Policy Optimization, Cross Entropy Method, Covariance Matrix Adaption Evolution Strategy, Deep Deterministic Policy Gradient.
• Fully compatible with OpenAI Gym.
• Includes support for running experiments on EC2 clusters and visualization tools.

Maintenance & Community

rllab is no longer under active development. It is maintained as garage by an alliance of university researchers.

Licensing & Compatibility

The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The project is no longer actively developed and recommends migrating to its successor, garage, for new projects and updates. The primary backend is Theano, which is also not under active development.