Research On Trajectory Planning Algorithms Of Autonomous Driving Vehicles

In the autonomous navigation of intelligent vehicles,trajectory planning is essential for driving safety and efficiency.The goal of trajectory planning is to generate an optimal trajectory for task accomplishment considering both physical and workspace constraints.The optimization consists of smoothness,risk,and efficiency terms.Generally,a trajectory consists of a path in Cartesian space and a speed profile in terms of executing time.Direct optimization of trajectory is usually non-convex and solved by numeri-cal optimization to find local optimums.Therefore,to reduce computation complexity,this thesis computes optimal path and speed profile separately.The path planning procedure is responsible for static obstacles in the environment while the speed planning procedure deals with dynamic ob-stacles.In the end,the spatio-temporal trajectory is computed by synthesizing reference path and speed profile.The main contributions of this thesis are as follows:· For path planning problem in structured and unstructured environments,a random sampling and numerical optimization based method is presented,which has a fast convergence speed and satisfies the kinematic and safety constraints.-Rapidly-exploring random trees(RRT)is an important approach in motion planning with a fast planning speed.Considering the low sampling efficiency and low convergence speed in conventional RRT methods,a heuristic sampling algorithm is proposed on the basis of the prior knowledge of the previous planning results and the potential field of the environment,which significantly increases the sampling efficiency.-Considering that the vehicle states and environments are varying,a pre-processing strat-egy for the RRT tree is presented.At the beginning of every searching period,the tree is initialized by the searching result of the last planning period.In this way,the searching efficiency in dynamic environments is increased.-Considering time limit,the searching result by RRT is usually not smooth enough.Therefore,for path smoothness and safety,a numerical optimization based method is adopted to optimize the path searched by the proposed RRT method in consideration of path length and safety.-For path planning in structured environments,the sampling and searching process is conducted in Frenet frame,which can represent lane constraints in on-road scenarios.In this way,the path planning algorithm can work well with structured environments.· For autonomous vehicles in dynamic environments,an RRT based speed planning method is presented.-For task formulation,the s-t motion space is constructed to describe the motion of the ego vehicle and obstacles.Then the high-dimensional trajectory space is mapped to the low-dimensional s-t space for computational efficiency.The speed optimization problem is transformed into the path searching problem in the s-t space,with designed constraints for collision avoidance and searching efficiency.-RRT based algorithm is proposed to search for the optimal speed profile in the s-t space asymptotically.The optimization indexes includes the comfort,safety and efficiency of the vehicles.-In each RRT searching step,node extension strategy is designed for the efficient ex-ploring of the s-t space with discretized time horizon;then the tree structure is locally refined for asymptotic optimization.The optimal speed profile is generated after the searching process converges.· A trajectory planning framework is presented for autonomous vehicles in dynamic environ-ments(structured and unstructured environments),which combines the path planning algo-rithm in Chapter 2 and speed profile plan ning algorithm in Chapter 3.Based on the overall trajectory optimization objective,the optimal trajectory is optimized to improve driving effi-ciency while guarantee safety.