System of obstacle avoidance used in deep-seabed vehicle based on BP neural network and genetic algorithm

A real-time obstacle avoidance neural network model was created according to the real working environment of deep-seabed vehicle. An information fusion method of multi-sensor was proposed to deal with the environmental situation collected by sonar sensors used in deep-seabed vehicle, and then the input of BP neural network was realized. Watching vector, steering angle and speed of the vehicle was set as output of BP network. Tutor training signal to achieve real-time obstacle avoidance was developed based on the moving way of both vehicle and man. Then, genetic algorithm was introduced to improve the BP neural network, so the problem of incidental trap in local minima with BP neural network was overcome. The simulation results indicate that the BP neural network improved by genetic algorithm can efficiently reach the expected target, and the problem of incidental trap in local minima is modified to a large extent. The simulation result of movement of the vehicle in the environment with obstacles shown in MATLAB indicates that the method is feasible.