Experimental evaluation of a vehicle steering assist controller using a driving simulator

Abstract

Human-in-the-loop driving simulator experiments are conducted to evaluate a proposed robust steering assist controller that is designed on the basis of driver uncertainty modelling. A nominal controller (NC) that is designed without consideration of driver model uncertainty is also tested for comparison. Two types of experiments are proposed: a long driving task with nominal configurations and a short driving task with initially large lateral position error. The data are analysed using both time domain and frequency domain metrics. In the time domain, the standard deviation of lateral position error and percentage of road departure are used. In the frequency domain, the stability margins and crossover frequency are used. The driving simulator results indicate that statistically, the designed robust controller shows improvements in the short driving experiments. The improvements in the long driving experiments are less evident because of driver adaptation. The non-robust NC suffers from high gain and should be avoided. The benefits of considering driver model uncertainty in the design of vehicle steering assist controllers are, therefore, justified.

Dependencies of driver steering control parameters

“…Drivers have different attitudes towards driving near the one lane boundary, being accounted for in this system. Chen and Ulsoy [7] describe a steering assist controller, designed to improve steering performance on the basis of driver uncertainty (degradation in performance) modelling. Kuriyagawa and Kageyama [8] have used two types of driver performance data to estimate the (elderly) driver state in relationship to potentially safety critical conditions, including tracking control parameters (preview length, steering gain).…”

A comparative analysis of electronic pedal algorithms using a driver-in-the-loop simulator and system identification of driver behavior

“…If driver maintains constant behavior regardless of the controller, a reasonable prediction of which pedal interpretation will work best could be made by examining the frequency response characteristics of the combined plant and controller. On the other hand, the famous "crossover model" of [8]- [10], which was supported by the results of many papers including [11], hypothesizes that the open loop transfer function of a driver cascaded with a plant has an invariant 20 dB/dec roll-off at crossover frequency, regardless of the plant. Additionally, preview/predictive models [12], which contain both a pursuit (feedforward) and compensatory (feedback) term, and do not necessarily produce results that coincide with those of the crossover model.…”

Vehicle Acceleration Prediction Using Specific Road Curvature Points

“…In the field of driving action description several clear-cut situations have been studied exhaustively: lane following (Fenton, 1988;Mammar et al, 2006), car following at a safe distance (Gipps, 1981;Olstam et al, 2004), lane change (Gipps, 1986;Salvucci et al, 2007). For lane following on a curved road an extensive theory has been developed, mainly based on control engineering approaches (Hsu et al, 1998;Yuhara et al, 2001;Chen et al, 2006;Mammar et al, 2006). Yet speed control (so called longitudinal control), including speed on curves, has only been studied extensively from a car stability perspective (Jin et al, 2007;Hel et al, 2007;Song, 2008).…”