CoGrid and Multi-User Gymnasium enable human-AI multi-agent research
Researchers introduce CoGrid, a multi-agent grid simulation library, and Multi-User Gymnasium (MUG), which converts simulations into interactive web-based experiments supporting arbitrary numbers of humans and AI agents.
Score breakdown
Researchers studying human-AI interaction and multi-agent systems can now deploy interactive experiments at scale without building custom infrastructure, accelerating empirical work on how humans collaborate with autonomous agents.
- 01CoGrid is a multi-agent grid-based simulation library with dual NumPy and JAX backends
- 02Multi-User Gymnasium (MUG) converts simulation environments into interactive web-based experiments
- 03MUG supports server-authoritative or peer-to-peer networking with rollback netcode for latency handling
Chase McDonald and Cleotilde Gonzalez address a gap in accessible tooling for human-AI multi-agent experimentation. They introduce CoGrid, a multi-agent grid-based simulation library offering dual NumPy and JAX backends for flexible computational backends. The second tool, Multi-User Gymnasium (MUG), bridges simulation and real-world research by translating CoGrid environments directly into interactive web-based experiments.
MUG's architecture supports arbitrary numbers of human and AI participants simultaneously, with flexible networking options: server-authoritative or peer-to-peer modes with rollback netcode to mitigate latency effects.
MUG's architecture supports arbitrary numbers of human and AI participants simultaneously, with flexible networking options: server-authoritative or peer-to-peer modes with rollback netcode to mitigate latency effects. Together, these tools enable researchers to deploy studies investigating human-AI interaction across psychology, cognition, and social decision-making. Both tools are released as open source with documentation and source code available at cogrid and multi-user-gymnasium readthedocs.io, and the paper includes case studies demonstrating their practical utility.
Key facts
- 01CoGrid is a multi-agent grid-based simulation library with dual NumPy and JAX backends
- 02Multi-User Gymnasium (MUG) converts simulation environments into interactive web-based experiments
- 03MUG supports server-authoritative or peer-to-peer networking with rollback netcode for latency handling
- 04Both tools are open source with documentation available at cogrid and multi-user-gymnasium readthedocs.io
- 05The tools enable research into human-AI interaction, social decision-making, psychology, and cognition