A high-performance engine for creating multi-agent reinforcement learning environments with thousands of pixel agents in gridworlds.
MAgent2 is a library for creating high-performance multi-agent reinforcement learning environments where thousands of pixel-based agents interact in gridworlds. It solves the problem of simulating large-scale agent populations efficiently, enabling research into artificial collective intelligence and competitive scenarios like battles. The project is a maintained fork of the original MAgent, offering reference environments built with the PettingZoo API for compatibility and ongoing support.
Researchers and developers working on multi-agent reinforcement learning, artificial collective intelligence, and large-scale agent simulations, particularly those needing environments with massive agent counts.
Developers choose MAgent2 for its optimized performance with very large numbers of agents, maintained codebase with regular updates, and seamless integration with the PettingZoo ecosystem, making it a reliable tool for cutting-edge multi-agent research.
An engine for high performance multi-agent environments with very large numbers of agents, along with a set of reference environments
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Optimized for efficiently simulating thousands of agents in gridworlds, enabling research on large-scale multi-agent systems as highlighted in the key features.
Reference environments use the standardized PettingZoo API, ensuring compatibility with reinforcement learning libraries and easing integration into existing workflows.
As a maintained fork, it receives regular updates, bug fixes, and support for new Python versions, providing reliability for long-term research projects.
Built for competitive scenarios like adversarial pursuits, making it ideal for studying artificial collective intelligence in conflict-based environments.
Only supports Linux and macOS, excluding Windows users, which restricts accessibility for some development and research teams.
Confined to 2D grid-based environments with pixel agents, lacking support for more complex spatial representations or continuous action spaces.
Contains only reference environments, requiring significant custom development for scenarios beyond the provided battle-focused examples.