Individual Driver Modeling via Optimal Selection of Steering Primitives

Abstract

A shared lane keeping assistance system supports the driver in the steering task. In contrast to an autonomous system, the shared system works in parallel with the driver by applying an additional torque on the steering wheel. This means the system and the driver perform the steering task in cooperation. As the driver is still part of the control loop a driver model is required to predict the steering behavior of the actual driver. In this paper we introduce a new lateral steering model which is suited to characterize individual drivers. This model describes, in contrast to other models, the neuromuscular system, limbs and its control for a specific driver by using a set of dynamic primitives (so called movemes). These movemes build a gray-box model for the neuromuscular system. The steering wheel angle is the predicted output of the moveme model. In order to generate a steering angle trajectory suited for the desired maneuver, the steering model switches between these movemes. Therefore, the central component of the driver model is a framework which determines the optimal switching sequence of the movemes. For this task an optimal control strategy is introduced. The approach is validated using a simulation of an ISO-double lane change with movemes which were identified from a set of real driver trajectories. The results show that the steering trajectories of the driver model highly correspond with the recorded driver trajectories.