MACAD-Gym provides OpenAI Gym-compatible environments for training Deep Reinforcement Learning agents in multi-agent connected autonomous driving scenarios. It targets researchers and developers working on autonomous driving systems, offering a flexible platform for simulating complex traffic interactions and testing RL algorithms.

How It Works

MACAD-Gym leverages the CARLA simulator as its backend, creating diverse driving scenarios. It abstracts CARLA's complexity into a familiar Gym interface, allowing seamless integration with various RL libraries. The platform supports homogeneous/heterogeneous agents, communication protocols, and customizable scenarios defined via JSON-like configurations, facilitating systematic benchmarking and research.

Quick Start & Requirements

• Install: pip install macad-gym
• Prerequisites: CARLA simulator (0.9.x or above), Python 3.x, Miniconda/Anaconda, cmake, zlib1g-dev. Ubuntu 18.04/20.04/22.04 or later recommended; Windows 10/11 requires CARLA Windows package and CARLA_SERVER environment variable.
• Setup: Requires CARLA installation and configuration.
• Docs: Learning Platform and Agent Interface, Getting Started

Highlighted Details

• OpenAI Gym-compatible interface for easy integration with RL libraries.
• Supports homogeneous/heterogeneous, communicating/non-communicating multi-agent settings.
• Scenarios are configurable via JSON-like files, allowing easy extension.
• Environments are versioned using a clear naming convention for benchmarking.

Maintenance & Community

The project is associated with the NeurIPS 2019 workshop paper. Contribution guidelines are provided.

Licensing & Compatibility

The repository does not explicitly state a license in the README. Compatibility with commercial or closed-source projects is not specified.

Limitations & Caveats

The project is tied to CARLA 0.9.x and above; a separate carla_gym environment exists for CARLA 0.8.x. The README does not detail specific hardware requirements beyond CARLA's own needs or provide performance benchmarks.