Data-driven human driver lateral control models for developing haptic-shared control advanced driver assist systems.

In order to improve road safety, many advanced driver assist systems (ADAS) have been developed to support human-decision making and reduce driver workload. Currently, the majority of ADAS employ a single, often very simple, driver model to predict human-driver interaction in the immediate future (e.g., next few seconds). However, there is tremendous variability in how each individual drives, necessitating personalized driver models, based on data collected from observed actual driver actions. Yet, because we currently lack sufficient knowledge of the high-level cognitive brain functions, traditional control-theoretic driver models have difficulty accurately predicting driver actions. Recently, machine-learning algorithms have been utilized to predict future driver control actions. We compare several of these algorithms used to predict the lateral control actions of human drivers. Specifically, we compare these algorithms in terms of their suitability to develop haptic-shared ADAS, which share the control force with the human driver. To this end, we need to know how the steering torque is provided by the driver. However, low-cost driving simulators typically measure steering angle but not steering torque. Thus, this work also proposes a methodology to estimate the steering-wheel torque. Using the estimated steering torque, we train several machine learning driver control models and compare the performance using both simulated and real human-driving data sets.