An AI AUV Enabling Vision-based Diver-following and Obstacle Avoidance with 3D-modeling Dataset

This paper presents an AUV with AI (artificial intelligence), which is able to perform real-time optical vision-based diver-following and forward looking altimeter-based obstacle avoidance. The AI AUV is equipped with thrusters and a standard navigation-related sensor suite. A diver detection convolutional neural network, a suite of motion controllers, and a diver detection payload device are developed to enable diver-following functionality of the AUV. An obstacle avoidance algorithm based on forward looking altimeters is developed to enhance the waypoint navigation security of the AUV with obstacle avoidance functionality. The diver-following and the obstacle avoidance capabilities of the Taiwan Moonshot AUV under different scenarios are evaluated through hardware-in-the-loop simulations. In addition, the designated single diver following capability of the Taiwan Moonshot AUV is also verified through closed water experiments conducted in a towing tank.