OGR-MARL: A New Framework for Multi-Agent Pursuit in Port Waterways
A new study introduces OGR-MARL, a framework for enhancing teamwork among different unmanned surface vehicles (USVs) in tight port waterways. This framework, detailed in arXiv:2608.12995, addresses the challenges of catching evasive targets while following navigation rules and traffic guidelines. OGR-MARL is versatile, working with various multi-agent reinforcement learning (MARL) algorithms by integrating shared beliefs about evaders, role-specific goals, and adaptive penalties. It refines actions based on guided behaviors rather than starting from scratch. The authors tested OGR-MARL with established continuous-control frameworks like MADDPG and MAPPO, resulting in variants like OGR-MADDPG and OGR-MAPPO. In simulations of the Xiazhimen port, OGR-MASAC achieved a 75.0% capture rate, showcasing its effectiveness.
Key facts
- OGR-MARL is a framework for heterogeneous USV cooperative pursuit in constrained port waterways.
- It is decoupled from specific MARL algorithms.
- Integrates shared evader belief, role-conditioned option targets, adaptive rule penalties, and residual policy learning.
- Instantiations include OGR-MADDPG, OGR-MATD3, OGR-MAPPO, and OGR-MASAC.
- Experiments in an abstract Xiazhimen port-waterway scenario.
- OGR-MASAC achieves a 75.0% capture rate.
- Paper available on arXiv with ID 2608.12995.
- Addresses navigation, traffic, and role constraints.
Entities
Institutions
- arXiv
Locations
- Xiazhimen