The invention relates to a path planning
obstacle avoidance control method for an autonomous
underwater vehicle in a large-scale continuous obstacle environment. The invention relates to the technicalfield of
underwater robot path
obstacle avoidance planning, and the method comprises the steps: building a large-scale continuous obstacle
simulation training environment, building the state and action of a deep
reinforcement learning neural network through employing
obstacle avoidance sensor information as the input and the navigation speed and
yaw angular speed as the output, for a multi-targetstructure of a
motion planning obstacle avoidance control process, performing
modular design on a reward function, and in order to avoid
system instability caused by sparse rewards, setting continuous rewards are set in combination with an artificial
potential field method. According to the method, obstacle avoidance training is carried out on the autonomous
underwater vehicle by utilizing an improved depth deterministic strategy gradient
algorithm, and an obstacle avoidance strategy obtained by training is written into a
robot lower
computer control system; when the autonomous
underwater vehicle runs in an underwater
canyon, obstacle avoidance is carried out by using an obstacle avoidance strategy learned by training, and the autonomous
underwater vehicle safely reaches a target area.