Zhu et al. Launch Open-Source Platform for AUV Swarm Tracking

Autonomous underwater vehicle (AUV) swarms face severe acoustic constraints, intermittent communication, and limited observations when tracking moving targets. Existing multi-agent reinforcement learning (MARL) solutions lack a unified open-source platform that incorporates realistic 6-degree-of-freedom (6-DOF) AUV dynamics. To fill this gap, Shengchao Zhu and colleagues from Hohai University and other institutions developed MARL-AUV—the first open-source platform connecting a public MARL training framework (DI-engine) with a physically modeled AUV swarm simulator. It provides a standardized experimental protocol for training, testing, and comparing representative RL and MARL algorithms under realistic underwater conditions.

On top of this platform, the team proposes STG-MAPPO (Semantic Task Graph-enhanced Multi-Agent Proximal Policy Optimization). STG-MAPPO builds semantic policy inputs from tracking diagnostics, task phases, observation confidence, link availability, neighbor tracking quality, and local role advantage. A compact semantic task graph links communication-constrained network states to decentralized actor decisions, while a velocity-level action abstraction maps high-level cooperative decisions to executable 6-DOF AUV controls. The work addresses key challenges in underwater target tracking, including changing communication topology, intermittent acoustic links, and limited per-AUV observation. Code is publicly available, enabling other researchers to benchmark and build upon this approach.