In 2013, Robert Wunderlich, director of the Texas A&M Transportation Institute’s (TTI’s) Center for Transportation Safety gathered a team of researchers to work on a project funded by the Toyota Economic Loss Settlement — money set aside for transportation safety research. The project’s goal was to find methods to reduce crashes caused by vehicle- and/or driver-based errors.
“The overall concept of the project was to see if we could use the technological power of computers in the car to monitor vehicle and driver performance,” explains Wunderlich.
To be successful, he knew, the three-year project would require a team of experts in technology, physiology, and modeling human and vehicle behaviors. The complex nature of the project required a search for experts outside TTI.
“TTI has extensive capabilities in this area, but we needed to find other specialists because the project crossed over so many different domains,” says TTI Human Factors Program Manager Mike Manser. “The researchers involved were all experts in their own fields, and that’s what made this project so successful.”
The multidisciplinary, multi-institute team Wunderlich assembled combined TTI’s infrastructure and traffic control expertise with the University of Michigan Transportation Research Institute’s (UMTRI’s) knowledge of vehicle operations and safety, Texas A&M University’s vehicle dynamics modeling and human factors expertise, and the University of Houston’s (UH’s) Computation Physiology Lab.
Along with Wunderlich and Manser from TTI, the team included:
- Reza Langari, head of the Texas A&M Department of Technology and Industrial Design;
- Tom Ferris, assistant professor in the Texas A&M Department of Environmental and Occupational Health;
- Ioannis Pavlidis, director of the UH Computational Physiology Laboratory and Eckhard Pfeiffer Professor of Computer Science; and
- Shan Bao, UMTRI associate research scientist.
“This project would not be possible without collaboration because it was an A to Z operation,” says Pavlidis. “Every link in this chain depended on the previous one, and each link was done by different university teams — if one failed, the project would never make it to a successful conclusion. Team science at its best!”
Detecting Driver Stress
To detect driver stress, researchers measured the heart rates, thermal body temperatures and perspiration rates of 68 volunteer drivers. The volunteers drove the same driver simulation course in four different scenarios — under normal conditions, while distracted with cognitively challenging questions, while distracted with emotionally charged questions, and while preoccupied with texting.
The driver simulation course was conducted at TTI, and data were then sent to Pavlidis at UH for analysis.
“The beautiful thing about data nowadays is that we can run the simulations in College Station and send them to Houston to be analyzed without any added travel costs,” Manser observes. “That speaks to the new way we do research these days — these collaborations can be done remotely and still be quite successful.”
Similar tests were also conducted using production vehicles on the RELLIS test track. The detection system was also tested on public roads in the College Station area.
Detecting Vehicle Error
Langari and his team at Texas A&M worked with the UMTRI team, led by Bao, and developed algorithms to determine vehicle-based errors or anomalies that might occur while driving — particularly unintended acceleration.
The UMTRI team analyzed naturalistic driving data, accumulated from a prior National Highway Traffic Safety Administration–supported study, for normal patterns of behavior so they could distinguish any abnormal behavior that would fall outside that range. The Texas A&M team produced their own human-subject data using a driving simulator to determine normal and abnormal driving and developed their own specific techniques for this purpose.
“Toward the end of the project, we jointly developed some test scenarios that were evaluated at Texas A&M in our driving simulator, and the data were shared with UMTRI,” explains Langari. “Both their methods and ours were applied to that same dataset to ensure we had alternative approaches to deal with this relatively complex problem.”
After three years of data collection and analysis, the team developed a countermeasure to detect vehicle-based malfunctions and driver-based errors.
“We were able to identify real-world problems — vehicle errors and high levels of driver stress — and conduct a series of projects to create a product that can be implemented immediately by vehicle manufacturers to improve safety,” explains Manser. “And it all came about through these series of experiments that started out with this very theoretical concept about what we could do.”