This repository provides a PyTorch implementation of the Multi-Agent Deep Deterministic Policy Gradients (MADDPG) algorithm, targeting researchers and practitioners in multi-agent reinforcement learning. It enables training agents in cooperative-competitive environments, as detailed in the MADDPG paper.

How It Works

The implementation follows the MADDPG algorithm, which extends DDPG to multi-agent settings. It utilizes an actor-critic architecture where each agent has its own actor and critic. The critic takes the observations and actions of all agents as input, allowing it to learn a centralized value function. This centralized critic aids in training decentralized actors, addressing the non-stationarity inherent in multi-agent RL.

Quick Start & Requirements

• Install the Multi Agent Particle Environment (MAPE) from https://github.com/openai/multiagent-particle-envs.
• Recommended PyTorch version: 1.4.0 (later versions may have issues with in-place operations).
• Clone this repository into the same directory as the MAPE.
• The main file requires the make_env function from the MAPE package.
• Tutorial video: https://youtu.be/tZTQ6S9PfkE

Highlighted Details

• PyTorch implementation of MADDPG.
• Designed for mixed cooperative-competitive environments.
• Based on the paper "Multi Agent Actor Critic for Mixed Cooperative-Competitive Environments" (https://arxiv.org/pdf/1706.02275.pdf).

Maintenance & Community

• The repository is a personal implementation by philtabor.
• No explicit community channels or roadmap are mentioned.

Licensing & Compatibility

• The repository does not explicitly state a license. The underlying MAPE environment has no explicit license mentioned in its README.

Limitations & Caveats

• Requires a specific, older PyTorch version (1.4.0) due to potential in-place operation issues in newer versions.
• Dependencies for the MAPE are described as "somewhat out of date," potentially requiring manual dependency management.
• The project appears to be a single author's implementation without extensive community support or ongoing maintenance signals.