Extended State Observer Based Linear Quadratic Regulator for the Path-Tracking of Self-driving Buses

Abstract

Smooth and accurate path tracking has a significant impact on safety and user-experience for self-driving buses. Path-tracking control, however, suffers from the uncertainties in the steering system, vehicle motion dynamics, and road-tire frictions, etc. In this paper, an extended state observer (ESO) based linear quadratic regulator (LQR) is proposed for the path-tracking of self-driving buses. In a departure from conventional LQR design, the bus motion dynamic is treated as a simple canonical model, the discrepancy of which from the real bus, caused by the nonlinearities and uncertainties, is lumped as a total disturbance state to be estimated by ESO online. With the total disturbance mitigated in the feedback loop in real-time, the bus motion dynamic is enforced to behave as the canonical model to be easily controlled by LQR. Experimental results show that relative to conventional LQR, the ESO-LQR design can enhance the transient response by 21%, and reduce the path-tracking error by up to 72 % in the presence of steering wheel angle offset.