This repository provides the code for the "Evolved Policy Gradients" paper, focusing on a novel approach to reinforcement learning policy optimization. It is intended for researchers and practitioners in deep reinforcement learning seeking advanced policy gradient methods.

How It Works

EPG introduces a method for learning policies by evolving a population of policies using evolutionary strategies. This approach aims to overcome limitations of traditional policy gradient methods by directly optimizing a population of policies, potentially leading to more robust and efficient exploration and learning.

Quick Start & Requirements

• Install: Requires Anaconda, mpi4py, gym[all], mujoco_py, and several other Python packages. Installation involves creating a Conda environment and pip installing dependencies.
• Prerequisites: Python 3.6.1, macOS, Open MPI, MuJoCo.
• Links: Paper: https://arxiv.org/abs/1802.04821, Demo: https://youtu.be/-Z-ieH6w0LA

Highlighted Details

• Implements Evolved Policy Gradients as described in the paper.
• Includes scripts for launching experiments, testing, and visualizing results.
• Leverages MPI for distributed computation.

Maintenance & Community

• Status: Archived (code provided as-is, no updates expected).
• No community links or active maintenance information provided.

Licensing & Compatibility

• License: Not explicitly stated in the README.
• Compatibility: Designed for macOS. Dependencies like mujoco_py may have specific licensing or installation requirements.

Limitations & Caveats

The project is archived and no longer maintained, meaning no updates or bug fixes are expected. The installation process is specific to macOS and older Python versions (3.6.1), and relies on several external dependencies that may be difficult to set up on modern systems.