Optimization of Ship Collision Avoidance Algorithms Under Uncertain Conditions of Target Ship Movement

Abstract The intelligent navigation system of maritime autonomous surface ships utilizes multi-sensor fusion technology to enable precise target positioning and real-time obstacle detection, these vessels often face persistent challenges in collision avoidance scenarios. For collision avoidance, these vessels often face difficulties in continuously sensing obstacle signals, rendering the understanding of the navigational situation and potential encounter situations challenging. Such perceptual limitations hinder the effectiveness of existing collision avoidance algorithms. Therefore, this study focuses on optimizing collision avoidance algorithms, particularly in scenarios where the navigation intent of obstacles is uncertain. To address the limitations of conventional velocity obstacle (VO) methods—such as handling long-term avoidance tasks and dense dynamic environments—we propose a hybrid algorithm combining dynamic window approach (DWA). This algorithm facilitates the creation of autonomous collision avoidance strategies for MASSs across various scenarios. Next, to tackle the problem of discontinuous output in the navigational state of the involved vessels, a convolutional neural network-gated recurrent unit (CNN–GRU) based network is proposed for vessel state prediction. This network uses historical AIS data to predict the involved motion state and their uncertainties, obtaining an uncertainty distribution of the vessel's position. This ensures that even when moving obstacles cannot be continuously sensed, the considered intelligent decision-making system can provide effective collision avoidance support for the vessel. Subsequently, the vessel's domain is optimized, taking into full account the uncertainty in its position. Finally, a complete collision avoidance decision-making system architecture is designed for MASSs.