This repository provides a PyTorch implementation of the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, designed for training agents in mixed cooperative-competitive environments. It is suitable for researchers and practitioners exploring multi-agent reinforcement learning strategies.

How It Works

The implementation follows the MADDPG framework, which extends DDPG to multi-agent settings by using a centralized critic that observes the actions and states of all agents, while each agent maintains its own decentralized actor. This approach allows for stable learning in complex, non-stationary environments where individual agent policies are constantly changing.

Quick Start & Requirements

• Primary install/run command: python main.py --help
• Prerequisites: PyTorch (0.3.0.post4), OpenAI Gym (0.9.4), Tensorboard (0.4.0rc3), Tensorboard-Pytorch (1.0). OpenAI baselines (commit hash: 98257ef8c9bd23a24a330731ae54ed086d9ce4a7) and a fork of Multi-agent Particle Environments are also mentioned.
• Links: OpenAI baselines, Multi-agent Particle Environments

Highlighted Details

• Implements MADDPG for cooperative-competitive scenarios like Physical Deception and Cooperative Communication.
• Includes a Predator-Prey environment with a single prey and three predators.
• Incorporates techniques like gradient norm clipping and policy regularization.

Maintenance & Community

• The project is a personal fork, with no explicit mention of active maintenance or community channels.

Licensing & Compatibility

• The README does not specify a license. Compatibility for commercial use or closed-source linking is not addressed.

Limitations & Caveats

The implementation is a personal fork and may not reflect the latest advancements or best practices. Features like ensemble training, inferring other agents' policies, and mixed continuous/discrete action spaces are explicitly noted as not implemented. The specified dependency versions are from the time of use and may not be strict requirements.