Planning and Control for Autonomous Vehicle Systems Based on Collaborative Neurodynamic Optimization

基於協作式神經動力優化的自主載具系統規劃與控制

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date3 Sep 2020

Abstract

Planning and control are two important modules for task completion in unmanned vehicle systems. Planning for autonomous vehicles includes task assignment and trajectory generation. Task assignment is to assign a group of vehicles to a group of tasks. Trajectory generation is to generate trajectories for the vehicles to destinations of specific tasks. Control of autonomous vehicles is to maneuver vehicles for tracking given trajectories from planning module. As complex unmanned tasks need cooperation to complete, and system dynamics of vehicles are usually nonlinear, the formulated planning and control problems are usually global optimization problems.

Neurodynamic approaches based on recurrent neural networks are suitable for various computation tasks, in particular, optimization tasks. Recently, collaborative neurodynamic optimization approaches are developed for global optimization problems by using multiple neurodynamic models computing in parallel for searching the global optima. Therefore, collaborative neurodynamic optimization is suitable for vehicle system planning and control formulated as global optimization problems.

The thesis consists of four parts for vehicle system planning and control in a unified framework. The first part aims to develop a task assignment approach for multivehicle systems via collaborative neurodynamic optimization. A collaborative neurodynamic approach is proposed for solving task assignment problems with zero-one constraints and cooperation constraints. In the second part, to solve the receding-horizon navigation planning problem formulated as sequential global optimization problems, a navigation planning approach based on collaborative neurodynamic optimization is proposed and proven to be convergent to complete trajectories of under-actuated vehicles. In the third part, to solve the receding-horizon trajectory generation planning problem for multiple autonomous surface vehicles, a distributed approach is developed based on collaborative neurodynamic optimization via interactive communications among neurodynamic models. In the fourth part, to control autonomous vehicles for trajectory tracking, model predictive control of under-actuated vehicles is formulated as sequential global optimization problems and solved by using a collaborative neurodynamic approach. In addition, the close-loop stability of vehicle systems is analyzed under given conditions and assumptions.