PRIMAL provides a distributed reinforcement and imitation learning framework for training multiple agents to collaboratively plan paths in 2D grid environments. It is designed for researchers and practitioners in multi-agent pathfinding (MAPF) and reinforcement learning, offering a solution for complex coordination challenges.

How It Works

The core of PRIMAL utilizes a distributed Actor-Critic approach, specifically A3C (Asynchronous Advantage Actor-Critic), adapted for multi-agent scenarios. It trains agents to learn policies that minimize collisions and reach goals efficiently. The framework includes a custom OpenAI Gym environment for MAPF, allowing for flexible scenario definition and agent interaction.

Quick Start & Requirements

• Install via pip install -r requirements.txt.
• Requires Python 3.4+, Tensorflow 1.3.1, Cython 0.28.4, OpenAI Gym 0.9.4, Numpy 1.13.3.
• C++ extension cpp_mstar requires compilation within the od_mstar3 directory.
• An online interactive demo is available for testing trained models.

Highlighted Details

• Supports both Reinforcement Learning (RL) and Imitation Learning (IL).
• Includes a custom Gym environment (mapf_gym.py) for MAPF.
• Provides scripts for environment generation (mapgenerator.py) and systematic testing (primal_testing.py).
• Offers an interactive visualization tool with hotkeys for scenario editing.

Maintenance & Community

• Authors: Guillaume Sartoretti, Justin Kerr.
• No explicit community links (Discord, Slack) or roadmap are provided in the README.

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 relies on older versions of Python (3.4) and Tensorflow (1.3.1), which may pose compatibility challenges with modern systems and libraries. The lack of explicit licensing information could be a concern for commercial adoption.