Research On Computational Optimal Control Methods For Automated Vehicle Motion Planning Problems With Complicated Constraints

The incessant development of automated vehicles has brought opportunities in promoting the traffic safety level,improving the transportation efficiency,and establishing intelligent transportation systems.Among the core technical modules in an automated driving system,motion planning is a critical aspect which directly determines the intelligence degree of the automated vehicle.Concretely,motion planning refers to computing a trajectory/path which is dynamically feasible for the vehicle,comfortable for the passengers,and collision-free from the detected obstacles.This work focuses on the automated vehicle motion planning scheme.The prevalent motion planning methods include the graph search methods,sample based methods,curve based methods,machine learning methods,and optimal control methods.Formulating the motion planning scheme as an optimal control problem is beneficial in being accurate,straightforward,objective and unified.However,the formulated optimal control problem is complicated because1)The vehicle kinematics is dynamic and nonlinear.2)The collision-avoidance constraints are highly nonlinear,even being non-convex and pathological.3)In terms of multi-vehicle motion planning schemes,the aforementioned two issues would accumulate.Particularly,the number of collision-avoidance constraints would grow geometrically with the number of cooperative vehicles.These factors are beyond the capability of the conventional methods in the optimal control community.As the high-performance processors and large-scale optimization methods develop,computational optimal control methodologies have gradually become prevalent due to their excellent capabilities in handling generic and complicated problems.In solving motion planning problems through computational optimal control methods,this study focuses on three critical issues:accurate model formulation,efficient offline solution,and fast online solution.In more details,the contributions and innovations lie in the following aspects:1)Automatic parking,cooperative lane change,and signal-free intersection management are selected as the applications,and the optimal control problems have been formulated accordingly.Precise collision-avoidance constraints among the rectangular vehicles are formulated in the analytical form,and the feasible boundary conditions for driving safety are proposed.2)The formulated optimal control problems are discretized as nonlinear programming(NLP)problems by the Orthogonal Collocation Direct Transcription method,and then solved via interior-point method(IPM).Two initialization methods are developed to facilitate the NLP-solving process when using IPM,namely the spatio-temporal decomposition method,and the progressively constrained dynamic optimization method.3)Two kinds of online motion planning approaches are proposed,which respectively utilize standard problem sets and simple problem transformation strategies to find near-optimal online solutions to optimal control problems.In handling online motion replanning problems,we have proposed a finite element reconfiguration based initialization method,and a multi-scale fault-tolerant optimization strategy.Simulation experiments have been conducted to verify the performances of the proposed methods in terms of solution capability,solution quality and computational speed.As the developments of sensing,communication,and control modules are becoming saturated,the enormous potential in the decision module would gradually attract attention.This research study focuses on how to derive high-quality motion planning solutions,and aims to provide a technical reference for the development of decision modules in the future automated vehicles.