Trajectory Optimization For Autonomous Vehicles Under Complicated Collision Avoidance Constraints

Autonomous vehicles have been developed rapidly recently. By fully grasping the real-time traffic information, accidents and casualties due to human error can be reduced by auto-vehicles adjusting their conditions timely. Many semi-auto driving techniques have been popularly applied in cars such as emergency braking, cruise controling, lane keeping etc. However, problems in auto-vehicles’ driving environment modeling under complicated traffic scenarios and collision-free trajectory optimization for multi-vehicles still cause much attention. This thesis focuses on issues including standard auto-parking trajectory planning with various parking slots, environment modeling under unexpected dynamic obstacles and multi vehicles’ collision-free cooperation under an intersection without traffic lights. These issues are studied based on a simultaneous approach.The main research contents and contributions are as follows:1.On the basis of the autonomous parking problem in the urban environment, a simultaneous framework is put forward for dynamic optimization for autonomous parking. Models of the vehicle as well as the environment with obstacles are handled all together in this approach. MPCC(mathematical programs with complementarity constraints) and R-functions methods are proposed to describe the conditional constraints relating to obstacle avoidance. Combined with the vehicles’ kinematic and physical constraints, a driving system model is established. Discretization based on orthogonal collocation over finite elements(OCFE) is applied to obtain accurate numerical solution of the dynamic problem. The resulting trajectory with time information can be employed directly to vehicle operation.2.Characterized with complicated environment constraints, the parking trajectory optimization problem sometimes may fail to converge, so a temporal and spatial segmentation algorithm is proposed to enhance the performance of the original algorithm. Attraction region and collapse region are introduced in the algorithm to partition the parking space. In this way, the original complicated environment constraints can be devided into simplified constraints in the fragmented space making an easier formulation to solve. Numeric results demonstrate the effectiveness of the new algorithm.3.Based on a complete map of urban driving environment, the global planning algorithm for multi-vehicles’trajectory optimization problem is studied. Combined with the complcated collision avoidance constraints between vehicles and dynamic obstacles, a multi-vehicle cooperating system model is established. Numeric results show the effectiveness of the simultaneous approach for global planning.4.When the surrounding environment is changeble, local rolling planning algorithm is studied under obstacle prediction model. The assumptive static model, assumptive speed-maintaining model and assumptive complete model are applied to predict the obstacle’s behaviour. Meanwhile, the vehicle system model is reformulated according to the relationship between the collision detection zone and the location of the obstacle. Numeric results of different prediction model are compared and some conclusions are put forward.