Modeling Driver Distraction from Cognitive Tasks

Modeling Driver Distraction from Cognitive Tasks Dario D. Salvucci (salvucci@mcs.drexel.edu) Department of Mathematics and Computer Science, Drexel University 3141 Chestnut St., Philadelphia, PA 19104 Abstract Driver distraction has become a critical area of study both for research in investigating human multitasking abilities and for practical purposes in developing and constraining new in-vehicle devices. This work utilizes an integrated- model approach to predict driver distraction from a primarily cognitive secondary task. It integrates existing models for a sentence-span task and driving task and investigates two methods in which the resulting model can perform multitasking. Model predictions correspond well qualitatively to two of three measures of driver performance as collected in a recent empirical study. The paper includes a discussion of the potential for building multitasking models in the framework of a production- system cognitive architecture. Introduction Computational cognitive modeling continues to mature rapidly as an area for both theoretical advances in understanding cognition and practical advances in developing intelligent technology. Cognitive modeling has grown from addressing only simple cognition in basic psychological tasks to capturing integrated cognitive, perceptual, and motor processes in large-scale complex, dynamic tasks (e.g., Chong, 1998; Jones et al., 1999). This paper investigates the application of cognitive models to an extremely common yet complex task: driving. Driving involves the continual multitasking of a number of subprocesses that make use of the driver's cognition, perception, and motor movements. This rich spectrum of necessary skills makes driving an ideal task in which to investigate how humans execute complex tasks and how computational models can represent and predict the multitasking behavior in these tasks. Driver Distraction and Cognitive Modeling One particular aspect of driver multitasking that has received enormous attention from media and researchers alike is that of “driver distraction” -- namely, the effects of multitasking while performing some secondary task. Numerous studies have now found that performing primarily perceptual-motor tasks while driving (e.g., dialing a cellular phone) can impair driver performance (e.g., Alm & Nilsson, 1995; McKnight & McKnight, 1993). These studies generally conclude, perhaps not surprisingly, that pulling a driver's visual attention from the road and/or her hand(s) off the steering wheel degrades the driver's ability to maintain a central lane position, follow a lead car with a constant headway, or react to potential hazards. Such studies have subsequently led to legislation at all government levels to restrict the use of potentially distracting secondary-task devices. While driver distraction is generally associated with effects on perceptual-motor processes, researchers have also reported that “cognitive distraction” arising from purely cognitive secondary tasks can adversely affect driver performance (e.g., Alm & Nilsson, 1995). These results are not fully conclusive and seem to depend highly on the secondary task as well as the driving situation; nevertheless, it is clear that even purely cognitive tasks can impact driver performance in certain situations. To better understand driver behavior and the sources of driver distraction, researchers have attempted to develop integrated driver models that capture driver behavior in a computational manner (e.g., Aasman, 1995). These models provide insight into the sources of distraction by elucidating the exact processes by which a driver attends to the external environment, processes this information cognitively, and then reacts and manipulates the environment. In addition, the computational models may be used to generate predictions about the effects of distraction on driver performance; for instance, the ACT-R driver model (Salvucci, Boer, & Liu, 2001) has been integrated with various models of cell-phone dialing to predict the impact of dialing on lane-keeping performance (Salvucci, 2001; Salvucci & Macuga, 2001). However, this previous work has addressed only primarily perceptual-motor secondary tasks with little cognitive component (like cell-phone dialing); to date, no models have demonstrated the ability to represent and generate “cognitive distraction” from primarily cognitive tasks. Modeling “Cognitive Distraction” This paper describes the first attempt to predict cognitive distraction with a computational cognitive model. This work employs the same methodology as in previous work for perceptual-motor distraction, namely the “integrated model approach” based in a cognitive architecture (see Salvucci, 2001). Cognitive architectures are computational frameworks that incorporate built-in, well-tested parameters and constraints on human cognitive and perceptual-motor abilities. This work focuses on a particular architecture, ACT-R (Anderson & Lebiere, 1998), that represents factual knowledge as declarative chunks and procedural knowledge as condition-action “production rules”. For our purposes, the ACT-R architecture has two important benefits: (1) it facilitates reuse and integration of multiple behavioral models, and (2) it provides built-in interfaces and default parameters that facilitate a priori predictions of real-world