NEURAL NETWORK-BASED PREDICTIVE STEERING CONTROL INTEGRATED WITH REAL-TIME COMPUTER VISION FOR SAFE AUTONOMOUS SHIP NAVIGATION

Abstract

The rapid advancement of autonomous maritime transportation has intensified the demand for intelligent steering control systems capable of ensuring accurate vessel maneuverability, reliable obstacle detection, and collision-free navigation under complex environmental conditions. This study presents an advanced neural network-based predictive steering control framework integrated with real-time computer vision for improving autonomous ship navigation, trajectory optimization, and operational safety. The proposed framework combines Convolutional Neural Networks (CNNs) for hierarchical visual feature extraction, Bidirectional Long Short-Term Memory (BiLSTM) networks for temporal vessel-motion modeling, and an attention-guided Model Predictive Control (MPC) strategy for adaptive rudder-angle prediction and dynamic trajectory regulation. The research considers a multimodal maritime navigation dataset comprising 125,000 synchronized observations collected across 2,500 simulated navigation scenarios, incorporating vessel heading, steering angles, ship velocity, GPS coordinates, maritime imagery, obstacle distances, wind disturbances, wave characteristics, and surrounding vessel movements. The experimental design encompasses open-sea navigation, congested maritime corridors, restricted visibility, sharp-turn maneuvers, and dynamic obstacle avoidance. The reported performance values indicate that the proposed CNN–BiLSTM–MPC architecture achieves 98.4% obstacle detection accuracy, 97.6% collision avoidance success, and 96.8% trajectory tracking accuracy. Furthermore, the framework attains a mean heading error of 1.24°, cross-track deviation of 0.38 m, and steering response time of 0.21 s, corresponding to improvements of 42.7%, 38.5%, and 35.6%, respectively, compared with conventional MPC-based steering control. The integrated visual perception module achieves a processing speed of 42 frames per second, supporting real-time navigation decisions while reducing estimated navigation-related energy consumption by 18.9%. Comparative evaluation against conventional Proportional–Integral–Derivative controllers, standalone MPC, and neural network-based steering architectures indicates improved navigation precision, disturbance rejection, and collision avoidance performance. These findings demonstrate the potential of integrating deep learning-driven visual perception with predictive steering control to advance intelligent maritime navigation, strengthen autonomous decision-making, and support the development of safer and more energy-efficient autonomous ship systems.