Real-time characterisation of driver steering behaviour

In recent years the application of driver steering models has extended from the off-line simulation environment to autonomous vehicles research and the support of driver assistance systems. For these new environments there is a need for the model to be adaptive in real time, so the supporting vehicle systems can react to changes in the driver, their driving style, mood and skill. This paper provides a novel means to meet these needs by combining a simple driver model with a single-track vehicle handling model in a parameter estimating filter – in this case, an unscented Kalman filter. Although the steering model is simple, a motion simulator study shows it is capable of characterising a range of driving styles and may also indicate the level of skill of the driver. The resulting filter is also efficient – comfortably operating faster than real time – and it requires only steer and speed measurements from the vehicle in addition to the reference path. Adaptation of the steer model parameters is demonstrated along with robustness of the filter to errors in initial conditions, using data from five test drivers in vehicle tests carried out on the open road.

Abbreviations: ADAS: advanced driver assistance systems; CG: centre of gravity; CAN: controller area network; EKF: extended Kalman filter; GPS: global positioning system; UKF: unscented Kalman filter