Research On Three-dimensional Path Following Methods Of Autonomous Underwater Vehicle

Underwater vehicle path following is a key technology for Autonomous Underwater Vehicle(AUV)to achieve automatic returning,underwater cruise,underwater terrain mapping,dynamic target strike and other operational tasks,and has been widely applied.In this paper,AUV mathematical model,three-dimensional path following and guidance method,backstepping sliding mode controller design method,design of controller based on deep reinforcement learning and improvement method,and construction method of autonomous underwater vehicle visual simulation system are studied to solve the path following and control problem of underwater autonomous underwater vehicle.Firstly,the relevant coordinate systems and motion model of path tracking are proposed,including geodetic coordinate system,carrier coordinate system and serret-frenet coordinate system.Kinematics model,dynamics model and path tracking errors model are proposed based on the architecture of AUV.In order to test the dynamic characteristics of the models and the ability to respond to the rudders,the maneuverability experiments of horizontal rudders and vertical rudders are carried out.Secondly,the virtual AUV set on the target path is used to guide the path following.Based on line of sight method(LOS),backstepping sliding mode technology and Lyapunov stability principle,a three-dimensional backstepping sliding mode controller for path following is designed.At the same time,a low pass filter is added to the control loop to suppress chattering in sliding mode control.Besides,multiple derivatives of multiple quantities exist in the control policy,in order to avoid the problem of differential explosion,a second-order filter is used to approximate the ideal virtual control signal.The three-dimensional path following of AUV is controlled by the angular velocity control policy and the expected velocity,and the performance of the designed path tracking controller are verified by simulation experiments.Thirdly,in order to enhance the compatibility of the path tracking controller,and simplify the controller design process,the controller is designed by using the deep reinforcement learning method.Several new deep reinforcement learning algorithms,such as DQN,Actor-Critic and Deep Deterministic Policy Gradient(DDPG)are compared,the characteristics,advantages and shortcomings of parameters update process are analyzed,and the DDPG algorithm is selected to design and train the three-dimensional path tracking controller.The training environment is built with Python and the simulation cycle is also established.Through the analysis of the training and simulation results,the design of the controller is improved,the reward function is updated,and the current disturbance observer is added to the control loop.The effectiveness is verified by retraining and simulation experiments.At last,the construction method of underwater path tracking visual simulation system is studied,AUV and other scene instances are modeled with Creator.The visual simulation system is established based on MFC/Vega Prime,and the interaction method of simulation scene and configuration method of special effects are studied.Through the comparison,the method of reading the motion simulation data directly from the file is selected,and it is displayed in real time in the visual simulation,and the DOF nodes are added to the AUV model to realize the real-time control of the AUV rudders.Finally,the visual simulation experiment of AUV three-dimensional path tracking is carried out to verify the effect of path tracking.