Towards COLREGs-aware ship collision avoidance with multi-agent PPO-LSTM in maritime IoT (opens in a new tab)
This paper proposes a three-layer maritime IoT framework for autonomous ship collision avoidance, integrating vessel states, risk metrics, and environmental data to comply with COLREGs (International Regulations for Preventing Collisions at Sea). It uses a multi-agent PPO-LSTM algorithm with a centralised-training-decentralised-execution approach to handle partial observability and complex vessel interactions. Simulations in congested and narrow waterways show improved success rates, safer distances, and fewer rule violations compared to other methods. Explicit COLREGs-aware rewards and timely communication are key to its performance.