Planning And Tracking Of Unmanned Vehicle Time-optimal Trajectory

Unmanned vehicles have a wide range of applications in the fields of transportation,firefighting and cleaning,however,the autonomous movement capability of unmanned vehicles is generally low.Based on this,researchers have proposed time-optimal trajectory planning for unmanned vehicles to ensure that they complete their tasks at the fastest speed.While trajectory planning often needs to consider factors such as safety,trajectory smoothness and planning efficiency,time-optimal trajectory planning needs to consider the optimization of both time and space dimensions,which is difficult to balance and more complicated to solve.To address the above problems,this paper designs and implements time-optimal trajectory planning for unmanned vehicles in known scenarios and investigates unmanned vehicle trajectory tracking.The main contents are as follows:Aiming at the time-consuming problem of calculating collision distance in trajectory planning,a strategy of segmented three-time spline trajectory optimization within a rectangular spatial corridor is proposed,in which the spatial corridor is generated by global path points obtained by Hybrid A* algorithm for expansion,and the trajectory optimization aims at obstacle avoidance and smoothness of trajectory for unmanned vehicles.Further for the problem of difficult time-optimal trajectory planning,this paper transforms the original nonlinear optimization problem into a discrete second-order cone programming problem,and obtains the trajectory by solving the Conic ALM(Augmented Lagrange Method).Finally,the traceability and time-optimal characteristics of the trajectory are verified by testing.For the nonlinear model prediction trajectory tracking controller input time delay problem,the Longacurta algorithm is proposed to be used to achieve an accurate estimation of the state.To improve the controller response speed,the prediction model and PHR-ALM(Powell-Hestenes-Rockafellar Augmented Lagrange Method)algorithm are proposed to be used for optimal control problem-solving.Finally,by comparing the simulation with the linear model predictive tracking algorithm,it is found that the improved algorithm has significant improvement in tracking accuracy and control output smoothness,but the response speed is relatively slow.Finally,this paper proposes an end-to-end trajectory tracking control method based on reinforcement learning to address the problems of low tracking accuracy and slow response speed of existing trajectory tracking methods.The trajectory tracking is described as a Markov decision process,a multi-task dense reward function is designed according to the tracking performance index,a state information and reference trajectory information encoding module is constructed,and it is connected to the Dueling DDQN(Double Deep Q-Network)policy network module.For the problem of effective feature extraction of local reference trajectory information,the attention mechanism is introduced in the network.The controller is obtained by simulation training.The algorithm is compared with model predictive control and pure pursuit control algorithms in the Gazebo simulation environment,and the results show that the proposed method has greater advantages in lateral error tracking and response time.