关键词: 行为基元, 轨迹跟踪控制, 粒子群优化, 微分博弈理论

Abstract: To solve the problem of unmanned vehicle trajectory tracking control with a sequence of human-like behavior primitives as the desired trajectory, a trajectory tracking control method combining offline optimization of behavior primitives and online game coordination is proposed. Based on the behavior primitive library extracted directly from real driving data, a model-based nonlinear optimization method is applied to generate a behavior primitive library that satisfies the constraints of vehicle kinematic properties. The optimal control parameters for each category of primitives in the behavior primitive library are obtained by offline optimization using the particle swarm algorithm, and a multilayer perceptual machine is applied to establish the mapping relationship between the optimal parameters of the controller and the categories of behavioral primitives. Based on the optimization of the control parameters within the primitives, the online game coordinated control method is used as the core to realize the optimal control parameter generation between the behavior primitives. The experimental results show that the proposed control method integrating behavior primitive optimization and gaming can significantly improve the tracking accuracy of the behavior primitive sequences and effectively solve the problem of stable and smooth transition between independent behavior primitives.

Key words: behavior primitive, trajectory tracking control, particle swarm optimization, differential game theory