MACRPO: Multi-agent cooperative recurrent policy optimization
2024

MACRPO: A New Method for Multi-Agent Cooperation in Reinforcement Learning

publication Evidence: high

Author Information

Author(s): Eshagh Kargar, Ville Kyrki

Primary Institution: Aalto University, Helsinki, Finland

Hypothesis

Can a new multi-agent reinforcement learning method improve cooperation and performance in environments without direct communication?

Conclusion

The MACRPO method shows superior performance in multi-agent environments compared to existing algorithms.

Supporting Evidence

  • The MACRPO algorithm was tested in three challenging environments and outperformed state-of-the-art methods.
  • Two novel mechanisms for information sharing were introduced, enhancing cooperation among agents.
  • Results showed that full cooperation among agents led to better performance in tightly coupled tasks.

Takeaway

This study introduces a new way for robots to work together without talking to each other, making them better at tasks like driving and navigating.

Methodology

The study proposes a new algorithm called MACRPO that uses a recurrent neural network to improve cooperation among agents in multi-agent environments.

Limitations

The method may struggle in high-dimensional discrete action spaces and could face scalability issues with many agents.

Digital Object Identifier (DOI)

10.3389/frobt.2024.1394209

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication