An intelligent autonomous surface vehicle manoeuvring analysis in restricted waters via neuroevolutionary method

Abstract

Autonomous surface vehicle (ASV) is beneficial to complete tasks that are beyond human limits such as offshore target detection, search-and-rescue (SAR) mission and offshore surveillance. The use of ASV may cater to lower the risks exposed to human operators in the shipping business for transporting cargo, handling congestion in waterways and harbours. Manoeuvring operations are complex and challenging for human pilots to conduct, in which with a slight error, may lead to marine incidents such as capsizing, grounding and collision. The application of ASV is advantageous as a human surrogate to operate in heavy traffic and restricted waters, therefore preserving human pilot safety and economic factors. ASV control is akin to the operation of human operators owing to the combination of mathematical models and artificial neural networks. This paper presented the capability of an ASV for decision-making to perform course-keeping and course-changing manoeuvring in restricted water. The neuroevolutionary method (combination single-layer artificial neural network (ANN) and genetic algorithm (GA)) is demonstrated in this work to utilize a simulation environment to train an ASV to operate without human intervention. The performance of the ASV to accomplish the piloting task faster compared to a human operator is highlighted as the result of the work. In this work, the movement of the ASV's rudder owing to an ANN bias due to imbalanced waterway turns setup (either to port or starboard direction). In future work, the training of the ASV may develop with implement the combination algorithms of multiple layer feedforward ANN and GA to perform autonomous navigation.