Fast Path Planning And Control Of Autonomous Vehicles With Complex Rood Environment

Autonomous vehicle is the product of multi-sensor sensing technology,computer science,intelligent control,high-speed communication network and high-performance processing equipment.It is one of the main carriers of artificial intelligence technology,which has great influence on the automobile industry,transportation industry and the whole social benefit.Therefore,it is imperative to investigate the key technologies of autonomous vehicles.This thesis studies the global path planning,local path planning and bottom control of autonomous vehicles.The main contents are listed as follows:First,a global path planning method for autonomous vehicles based on glowworm swarm optimization(GSO)in the traffic network topology is provided.A traffic network topology map is plotted according to the actual road traffic distribution.Then multiple feasible paths are generated on the topology map,and several sub-optimal paths are selected according to the GSO.Later on,the optimal path in the current traffic environment by switching adjustment optimization algorithm is suggested.A fast secondary planning algorithm is invoked if impassable situation occurs,which applies the previous searching results to implement fast global path planning under the current traffic condition.Second,the idea of model predictive control is utilized for local path planning.In the context of real-time change of the surrounding environment and the constraint conditions of the vehicle,the four times polynomial fitting,curvature solution and convexity discrimination of lane line are proposed.The operations of straight road maintenance,cornering lane deceleration,lane-changing overtaking,vehicle following and main lane obstacle avoidance are achieved.In addition,the detailed procedure of multi-circle approximating the vehicle and the obstacle is formulated,which simplifies the calculation of the shortest distance among the main vehicle and the surrounding objects.Next,a flexible combination scheme of feed-forward control and model predictive control(MPC)is proposed.The feed-forward control acquires the control variable corresponding to the change of the heading angle at different reference points in the front path.The particle swarm optimization(PSO)algorithm is utilized during the optimization process of MPC.Thus,the controlled quantity for each predictive horizon obtained in the feed-forward control is taken as the reference point of the PSO algorithm.By setting the reference point as the origin and choosing a suitable radius,a new optimization solution domain is obtained.PSO is used to optimize in this new region.Due to the decrease of the search range,the number of particles and the maximum number of iterations can be greatly reduced.As a result,the optimization time is shortened on the premise of the desirable control accuracy.Finally,in order to solve the model predictive control with constraints more accurately and faster,the generalized Lagrange multiplier method is used to construct the generalized cost function.The constrained problem is transformed into an unconstrained one.In addition,the quantum particle swarm optimization(QPSO)with a stronger global search capability is performed to optimize the solution,and the QPSO is designed in parallel so as to achieve an efficient and accurate optimization solution.