Monte Carlo Methods for Real-Time Driver Workload Estimation Using a Cognitive Architecture

Abstract

Human-machine interaction gets more and more cooperative in the sense that machines execute many automated tasks and cooperate with the human operator, who also performs tasks. Often some tasks can be executed by both, like a car that can autonomously keep the lane or is steered actively by the driver. This enables the human machine system to dynamically adapt the task sharing between machine and operator in order to optimally balance the workload for the human operator. A prerequisite for this is the ability to assess the workload of the operator in real-time in an unobtrusive way. We present two Monte Carlo methods for estimating workload of a driver in real-time, based on a driver model developed in a cognitive architecture. The first method that we present is a simple Monte Carlo simulation that gets as input the information that the driver can perceive, but does not take the actions of the driver into account. We evaluate it based on a driving simulator study and compare the workload estimates with functional near-infrared spectroscopy (fNIRS) data recorded during the study. Afterwards the shortcomings of the simple approach are discussed and an improved version based on a particle filter is described that takes the driver's action into account.