Research On Attitude Control And Path Planning Of Underwater Vehicle Based On Visual Ranging

With the advancement of marine scientific research and the gradual deepening of human exploration of the ocean,underwater vehicle,with their powerful environmental awareness and underwater operation capabilities,are playing an important role in areas such as seabed topographic mapping,underwater platform maintenance,and so on.Accurate attitude control and path planning are the basis for autonomous underwater vehicle execution.The traditional path planning method first finds the optimal path,then drives the underwater vehicle along the set path.Due to the disturbance of current,water pressure and other factors,additional attitude adjustment during the course of the underwater vehicle may deviate from the established route,which may affect the performance of the actual task.To address the above issues,this paper carries out the research of underwater vehicle attitude control and path planning algorithm based on visual ranging,achieving synchronous path and attitude planning for underwater vehicle,and further expanding the research results to maintain the formation of multiple underwater vehicles.The main research content is as follows:To solve the problem that it is difficult for the underwater vehicle to maintain its posture stability in the visual observation task,an algorithm for attitude control and path planning of the underwater vehicle based on visual distance measurement is presented.By using the position and attitude of the underwater vehicle and the binocular visual ranging method,the outline of the target is fitted linearly,and the relationship between the line of sight and the target is established to guide the underwater vehicle to adjust its attitude.A synchronous attitude control and path planning algorithm based on the improved Twin Delayed Deep Deterministic policy gradient(TD3)is designed,and a reward clipping strategy is proposed to optimize the output of the reward function so that the state with higher learning value has greater reward potential.Adding a random state expands the scope of exploration,fosters diversity of training samples,and further improves the generalization ability of the algorithm.And layer normalization(LN)is used to prevent the gradient disappearance in the neural network to stabilize network output and accelerate convergence.Finally,the effectiveness of the proposed algorithm was verified through simulation and indoor pool experiments.An improved TD3-based distributed algorithm for formation keeping and path planning is designed to solve the attitude control problem of multiple autonomous underwater vehicles with constrained sight range and attitude.Based on the leaderfollower formation topology and dynamic error model,the leader route planning controller and follower route tracking controller are designed to achieve route planning and formation keeping.In addition,in the early stages of training,expert experience is introduced to pre-train the network of leader.In order to reduce the impact on follower controller value evaluation process,follower network is delayed to update to ensure convergence.Finally,the effectiveness of the proposed algorithm is verified by simulation.