Preview-scheduled steering assistance control for co-piloting vehicle: a human-like methodology

As a significant active safety technology towards fully autonomous driving, driver-automation co-piloting provides a new thinking for enhancing vehicle active safety and relieving driver's workload. However, the preview pattern divergence between driver and steering assistance system (SAS) may give rise to serious steering conflicts, which reduces driver's faith in SAS, cause him/her to turn off it, and even result in fatal traffic accidents. Focusing on mimicking the driver's steering behaviour, this paper proposes a human-like steering assistance scheme by adaptively varying the preview time and assistance gain of the SAS controller. Driver's preview behaviour of the road curvature is numerically identified using simulator tests along a multi-curves road, based on which a nominal preview-scheduled driver model is established. Based on the nominal driver model, a co-piloting system featuring human-like preview behaviour and a time-varying assistance penalty is proposed to formulate the steering torque-overlay dynamics of both agents in path-tracking. Next, a linear parameter varying (LPV)/H infinity controller satisfying pole placement is integrated to ensure the robustness and stability within the entire parameter space. This enables the steering assistance scheme to have a scheduled preview pattern like a human driver. Results of driving simulator experiment indicate that the proposed steering assistance scheme is able to mitigate driver-SAS conflicts and improve driver's path-tracking performance.