A reinforcement learning backstepping-based control design for a full vehicle active Macpherson suspension system

Abstract

This paper presents a reinforcement learning backstepping-based control scheme for the design of a full vehicle active suspension system such that both the superior ride comfort and stabilization are achieved. It is well known that the traditional backstepping control scheme is suitable to solve higher order nonlinear system based on the strict-feedback form; however, the disadvantage of the procedure requires computing the derivatives of virtual control signals, which is a very complex task. Therefore, a reinforcement learning scheme using the deep deterministic policy gradient (DDPG) control strategy is developed to replace the process of finding virtual control force in backstepping method. It not only avoids the complexity of analytically calculating the derivatives of the virtual control signals, but also retains the systems robustness when there exist random disturbances from road irregularities. To verify the performance of the proposed active suspension control scheme, the random road unevenness profiles according to ISO 8608 is considered as vertical and lateral disturbance input excitation of a full-vehicle suspension system. Compared with conventional passive suspension system and backstepping control scheme, the proposed approach can be demonstrated to not only effectively improve ride comfort under random road excitation, but also improve the transient response and robustness.