Rear-end collisions account for approximately 28% of all crashes, resulting in 157 million vehicle hours of delay annually, or roughly one third of all crash-caused delays (National Safety Council, 1996). Driver inattention has been identified as a contributing factor in over 60% of these crashes (Knipling et al., 1993). Because inattention is such a powerful contributor to rear-end collisions, rear-end collision avoidance systems (RECASs) may help resolve this problem.
Understanding how to mitigate dangers associated with inattention and distraction is becoming increasingly important because emerging technology has the potential to increase driver distraction (Lee, Caven, Haake, & Brown, 2001; Mollenhauer, Hulse, Dingus, Jahns, & Carney, 1997; Parkes, 1993). The possibility of increasing driving safety using RECAS has generated a substantial body of research (An & Harris, 1996; Dingus, McGehee, & Hankey, 1997; Hirst & Graham, 1997; Knipling et al., 1993). Although several RECASs are currently in development, substantial uncertainty exists regarding driver response to these systems and the effects they will have on driving safety (McGehee & Brown, 1998; Tijerina, 1998).
An important component of collision avoidance systems is the warning algorithm, which determines the timing of the warning; consequently, its design is as important as the design of the driver interface. A poorly timed warning may actually undermine driver safety (McGehee & Brown, 1998). An alert issued too early may be ignored by drivers if they are unable to perceive the cause of the warning. If an alert occurs too late, drivers may view it as ineffective, and a late alert may even disrupt an ongoing braking process. Understanding the influence of alert timing on driver response is crucial to estimating collision warning effectiveness.
The type of automation that a collision warning represents provides a theoretical basis for examining how alert timing might influence driver performance (Parasuraman, Sheridan, & Wickens, 2000). Although the type of automation is primarily dependent on its design, it also depends on the users' interpretation of the automation. For example, drivers might consider a collision warning system to be automation that acquires and analyzes information to specify an appropriate driver response (automation that triggers a response). Alternatively, drivers may view the same system as automation that simply detects abnormal events and alerts the driver (automation that redirects attention). 'Whether drivers view the collision warning system as automation that triggers a response or as automation that redirects attention can have important implications for design (Parasuraman et al., 2000).
A collision warning system that triggers a response is likely to generate an open-loop response that could neglect important considerations regarding surrounding traffic and the alert validity (Lee, Gore, & Campbell, 1999). In this situation, the false warning associated with algorithms that provide early alerts could undermine safety by triggering an inappropriate braking response. Viewed as automation that redirects attention, the collision warning system is likely to generate a closed-loop response. If this is the case, then braking is modulated according to how the traffic situation impinges on the driver's perception of the field of safe travel (Brown, Lee, & McGehee, 2000; Gibson & Crooks, 1938). If this is the case, then late alerts would provide a smaller safety benefit because drivers may not have enough time to interpret the driving situation and generate an appropriate response. These theoretical distinctions have important implications for algorithm design. If the collision warning system acts as automation that triggers a response, then a relatively late warning that minimizes false warnings might be the best design alternative. …