The ATLAS Center partners embrace the power of competition to produce the highest quality research results. Each institution uses a competitive selection process. Experience has shown that Principal Investigators (PIs) who participate in and are selected from a competitive process are best equipped to produce high-level, peer-reviewed research that is well-organized, productive, and timely. Each partner conducts a competition for research funds using a selection process that follows a peer review process.
Associate Research Scientist, UMTRI Biosciences Group & Research Director, CMISST
Co-Investigators: Michael Elliott (UM School of Public Health and Institute for Social Research), Kaye Sullivan (Ford Motor Company)
Project Title: Development of a Statistical Method for Predicting Human Driver Decisions and a Paradigm for Communicating Predicted Decisions to Automated Vehicles in the Connected Environment
Abstract: As we move towards a future of vehicle automation, there will be a long transition period in which the driving environment will include a mix of automated and non-automated vehicles. While automated vehicles can communicate future trajectories and speeds to each other, human drivers do not communicate their intent. Being able to predict what a human driver is planning to do can substantially aid automated vehicles in planning and even improve the performance of driver assistance systems. We propose to use functional data analysis approaches to develop predictive models of driver decision-making in three contexts: 1) left turn go/stop-then-go, stop/no-stop at stop signs, and yellow-light go/stop. We will use existing driving data to identify a sample of these scenarios. Predictors will be developed from kinematic variables that can be measured using vehicle-based sensors. These may include speed, lateral and longitudinal acceleration, pedal use, position in lane, and following distance. Key outcomes of this effort include: 1) A general statistical modeling approach for handling prediction of driver decisions based entirely on sensor data; 2) A paradigm for conveying predictions and prediction uncertainty to an automated vehicle in a vehicle-communication setting; and 3) Prediction models of driver decision for the scenarios analyzed. This work can be generalized to a variety of driver decisions and has wide applicability to improve driver assistance systems and the performance of automated vehicles co-existing in an environment with human-controlled vehicles.
Research Professor, UMTRI Driver Interface Group
Co-Investigators: Brian Lin (UMTRI), Paul Milgram (University of Toronto), David Mitropoulos-Rundus (Hyundai Kia)
Project Title: Predicting Performance with Touch Screens to Determine Compliance with the NHTSA Distraction Guidelines
Abstract: NHTSA’s visual-manual distraction guidelines specify use of the visual occlusion method, whereby subjects wear a pair of goggles that cycle between being open (seeing the interface) and closed (not seeing the interface but seeing the road) to determine the task time. In an ongoing ATLAS project, this team is improving Pettitt’s method, a simple calculation that predicts total occlusion task times using standard times for elements in SAE Recommended Practice J2365, such for as pressing an alphabetic button.
Unfortunately, J2365 lacks times for touch screen gestures, gestures common to contemporary interfaces. To obtain them and validate J2365, a recently completed experiment (Lin, Green,Kang, Mize, Best, and Su, 2012 – 24 subjects, 14 tasks) involving a pre-production interface will be re-analyzed. Fortunately, the experiment log contains the time of each screen contact and screen update, readily be combined with the video record (more than 24 hours) of what subjects did. Using the prediction equations, the differences between predicted and measured task times, and regression analysis, the computational method and element times will be examined.
Subsequently, new standard times will be developed directly from the data each element or prospective element, including touch screen gestures. Using these improvements, new total task time estimates will be calculated and compared the experimental data.
The results will be summarized in (1) a revised version of SAE J2365, (2) a technical report and (3) and a journal article, and (4) presented to the SAE Safety and Human Factors Committee.
Thus, this research will be conducted by a team highly experienced with visual occlusion research, using existing data, to complete development of a low-cost method codified as an SAE Recommended Practice (being revised by the Project Director), concerning a DOT-identified high priority, safety-related topic. This procedure represents a fundamental change in driver interface evaluation.