Considering several issues faced in the design of autonomous vehicles, such as poor adaptability to the environment and traditional path tracking control algorithms that require manual optimisation of the control parameters, this paper proposes a mixed kernel-based dual heuristic programming (MKDHP) path-tracking control method. First, an error state equation is established based on the autonomous vehicle and the constructed road model, and a Markov decision process model is derived for the autonomous vehicle path tracking problem. Second, a return function is designed based on comprehensive error performance, safety, and stability indicators. A mixed kernel function is used to construct the basis function, and the recursive least-squares temporal difference (RLS-TD) algorithm is used to learn the critic. Finally, the path-tracking performance of the MKDHP method is verified through numerical simulations and hardware-in-the-loop experiments. The controller based on the MKDHP algorithm achieves maximum lateral errors of 0.37 and 0.42 m under varying road adhesion coefficients during urban road tracking, with average lateral errors approaching zero and root mean square errors of 0.06 and 0.04 m, respectively. This method does not require repeated debugging of control parameters to achieve favourable control accuracy on roads of different shapes under different vehicle speed conditions, thereby improving the environmental adaptability and autonomous optimisation ability of autonomous vehicle path tracking control methods.
Zhang et al. (Wed,) studied this question.