Local Traj Ectory Planning Of Autonomous Vehicles In Structured Environment

Autonomous driving is a highly complex system composed of multiple subsystems,including algorithms for positioning,sensing,decision-making,planning and control.Trajectory planning is an important part of it,ensuring the safety and efficiency of autonomous vehicles.Local trajectory planning generally follows the results of global path planning.Based on the current vehicle state and real-time environmental information(other road participants or static obstacles in the map),real-time planning is required to meet vehicle movement constraints and obstacle avoidance con-straints in the future.The trajectory has spatial and temporal information.Solving two spaces at the same time leads to higher dimensions and greater computational complexity.Therefore,this paper decouples trajectory planning into two sub-problems:path planning and speed planning,which re-duces the dimension of the solution and can ensure the real-time performance of the calculation.This article considers avoiding static obstacles in the path planning,gives local target points according to behavioral decisions,and plans the corresponding paths.In speed planning,avoidance for dynamic obstacles and the interaction of the vehicle with other road participants are considered.Finally,the planned path and velocity curve are synthesized into a local trajectory.The main research results of this paper are as follows:·Aiming at the path planning problem in a structured environment,using a sampling search strategy based on dynamic planning,a path planning algorithm that satisfies real-time and vehicle motion constraints is proposed:-In order to ensure real-time performance,a priori knowledge of a structured environ-ment(road)is fully utilized,and the road centerline is used as a reference line,and sampling is performed along this reference line to limit the search space to a reasonable range and improve the efficiency of the algorithm.-Using the Frenet coordinate system,the two-dimensional path planning problem is transformed into a one-dimensional problem in S-L space,which reduces the solution dimension,and through dynamic programming,the path complexity is further reduced.-Consider the vehicle dynamics model at the planning layer,perform lateral dynamics modeling,and transform it into curvature constraints based on the current state of the vehicle,so that the planning results conform to the current vehicle motion constraints,ensure the enforceability of the path,and reduce the burden of the underlying control·Considering the obstacle avoidance needs of other road participants,a speed plan based on Markov decision-making is proposed,which considers the behavior interaction between itself and other road participants to improve traffic efficiency.-Model the speed planning process as a partially observable Markov decision process.Use Monte Carlo tree search to perform real-time online planning and solve incomplete problems of offline strategies.Uncertain predictions of other road participants Per-formance and interactive modeling with the vehicle trajectory come in,which is more efficient than reactive speed planning.-Discrete the state and action to solve in real time,and obtain a smooth and continuous st curve by processing the optimal acceleration sequence using spline interpolation and integration processing.Under the premise of improving the traffic efficiency,the ride comfort is guaranteed.·Aiming at the problem of local trajectory planning,a physical simulation platform based on V-REP and ROS was established.Model predictive control was used as the trajectory tracking control algorithm.In combination with the proposed local trajectory planning algorithm,a real-time autonomous vehicle planning control system was established.In the system,the effectiveness of the algorithm is verified.