The Science of Driving
At present there is no comprehensive and accepted scientific understanding of how people drive cars. The current lack of a true predictive capability for driving is a clear weak point in delivering safety and efficiency benefits in the future.
UMTRI recognizes the Science of Driving (SciD) as a strategic "must" for the future of research in highway transportation. The goal is difficult - knowledge is needed from different disciplines, and even there the knowledge is not mature; so new and coordinated research is required to fill in the significant gaps in previous research.
The SciD initiative is based on three simple aims:
- To discover an enduring set of fundamental scientific principles, concepts, and theories of the driving process
- To develop a body of applied knowledge for understanding and predicting the driving process
- To address challenging research problems with first-order quantitative descriptions of "what's important?" and "what happens?"
The intent is to promote sound conceptual principles to accelerate progress in new areas of applied research for which the dynamics of the driving process is a key factor. Relevant application areas include: integrated vehicle controls, driver assistance systems, communication systems in vehicles, road geometry design, vision enhancement, designing systems to address driver impairment and reduce distraction. Any aspect of system design related to driving will see great benefit, with SciD reducing the search space for new and effective designs.
SciD is currently focused on four themes:
- How should we share driving control with smart technologies?
There is a large gap between the 'simple' driving modes of purely manual and purely automated driving. New technologies are starting to move driving into that gap, with systems like Adaptive Cruise Control. What is needed to make this trend safe, predictable and effective? - Understanding conflicts and measuring crash risk
Crashes often occur for complex reasons, involving the precise timing of decisions and control actions between two or more drivers. Detailed predictions of driver intent and actions can be simulated to predict outcome probabilities over time periods in the region of one-to-ten seconds. The relevant traffic interactions are closed-loop and involve driver perception and anticipation as much as they involve physical vehicle trajectories. Though existing traffic simulation models do predict certain types of traffic interactions, there are no such models available today to perform simulations relevant to emergent crash risk. - Estimation of driver information processing
The driver is essentially a "black box" for transportation researchers. Although some attempts are now being made to use brain imaging technology to look inside that box, other techniques are needed. In particular it is crucial to be able to estimate the time sequence of perception, anticipation and control states of the driver, just using available information relating to eye glance, external scene and driver control actions. A model-based approach becomes feasible once suitably detailed and comprehensive driver models are available to SciD. - How do people learn to drive?
Learning to drive is a fundamental aspect of driving behavior, and is more than simply the acquisition of skill - it also includes a significant experiential component for making tactical and strategic decisions. A model-based approach should form part of the SciD model development, and output should explain known facts, such as the rapid decrease in crash rates during the first six months of driving.
Ultimately, UMTRI sees these efforts taking place on an international scale, and some steps have already been undertaken to initiate a strategic approach to developing the core driver models. In addition to existing driver modeling efforts - within UMTRI, and in collaboration with U-M faculty - a key credential for Michigan is the naturalistic driving data already available within UMTRI databases.
