Path planning of Unmanned Surface Vehicle based on convolutional and long short-term memory neural network

Outdated planning methods of unmanned surface vehicle (USV) have many bottlenecks such as sub-optimization of paths for the lack of information, poor re-planning performance and inability to incorporate International Regulations for Preventing Collisions at Sea (COLREGS). This article emphasizes the deep learning algorithms applied to path planning. We innovatively build a CNN-LSTM(C-LSTM)-based path planning algorithm framework. The Convolutional Neural Network (CNN) and the Long Short-Term Memory (LSTM)-based planning algorithm have the capability to excavate the image space and time series, respectively. Firstly, the hierarchical structure is employed to handle environmental information, and combine vehicle-obstacles speed, heading and other information to vectorize the area of obstacles. Secondly, a dynamic obstacle with consideration of COLLEGS is modelled to improve the obstacle-avoidance ability and the decision-making ability combined with regulations. By comparison, the C-LSTM algorithm has higher path generation capabilities than existing algorithms by the extensive simulation results.