$2.5M U-M Research Initiative Will Analyze Data from Connected Cars
In an effort to tackle the sea change associated with the future of transportation, two University of Michigan research projects recently scored $2.5 million in grants from the Michigan Institute for Data Science (MIDAS) to analyze the reams of data being generated by autonomous and connected vehicles across the country.
MIDAS awarded funding to the two projects under the first round of its Challenge Initiative program, with another $120,000 contributed to each project by U-M’s Dearborn campus. The funding is part of U-M’s $100 million Data Science Initiative announced last fall.
U-M College of Engineering professor Pascal Van Hentenryck will lead “Reinventing Public Urban Transportation and Mobility,” a project to help design and operate an on-demand public transportation system for urban areas using predictive models, in which a fleet of connected and automated vehicles are synchronized with buses and light rail. The goal is to begin testing at U-M within a year, and then expand the experiment to Ann Arbor and Detroit.
The other project, “Building a Transportation Data Ecosystem,” will be led by the U-M Transportation Research Institute’s (UMTRI) Carol Flannagan. It will focus on creating a system that allows researchers to access massive, integrated datasets on transportation in a high-performance computing environment.
“We’re trying to build the pieces needed to put together a well-integrated, high-performance system for transportation data,” Flannagan said. Her project includes researchers from the School of Public Health; College of Engineering; College of Literature, Science and the Arts; UMTRI; and the Institute for Social Research.
Flannagan’s project involves three main tasks: developing a faster, more optimized computing infrastructure; working with biostatisticians to incorporate statistical methods in order to advance the analysis of transportation data; and identifying and addressing patterns in the data. The overarching goal of the project is to create a common repository of data—including traffic, weather, accidents, vehicle messages, traffic signals, and road characteristics—that will shape the development of connected and automated vehicle systems of the future.
“A lot of work being done with autonomous vehicles focuses on feature detection,” she said. “And all of that needs data infrastructure to form a base that enables analysis and applications. When we make the data accessible, then other people can get at it and expand opportunities and ideas.”
Flannagan said the raw data that researchers will study comes from pilots run by UMTRI where volunteers are given connected vehicles to drive for a number of weeks. “We collect speed, acceleration, braking, and steering data,” she added. “And sometimes there’s video.”
The datasets contain “huge volumes of messages,” Flannagan pointed out, and warning signals that can tell when drivers braked hard, for instance, or faced other safety issues. More than 20 states are contributing data, allowing researchers to link crash datasets with driving datasets based on location and time to see whether congestion, weather, or vehicle type had any effect on safety. Researchers are interested in getting their hands on as many datasets as possible without violating privacy protections, she said.
“We’re really excited that the state and auto manufacturers are embracing and investing in this kind of research,” Flannagan said. “The idea that we’ll continue to build these technologies in Michigan is really great.”
Van Hentenryck said transportation mobility is the best predictor of social mobility—more so even than crime or quality of public schools—which is why he considers his project to be key to improving the delivery of public transportation.
“It’s also important for health care,” he said of increasing transportation mobility. “Research has shown that 3.6 million people in this country each year don’t seek healthcare because they don’t have access to transportation. Twenty-three million people don’t have access to a supermarket without transportation. Mobility is really important in the United States. If we make improvements, we can solve some of these issues.”
Van Hentenryck’s research project will look at how the transportation infrastructure is used and how connected and autonomous vehicles are incorporated into new solutions, especially the challenge of getting people from their homes or to final destinations in the transit system. His collaborators on the project include researchers from the schools of Information, Engineering, Emergency Medicine, Architecture and Urban Planning, and Computer Science.
“If we can design new models using on-demand transportation—a shuttle that picks you up and drops you off at the light rail station, for example—we can connect people to the infrastructure so they can use transit more effectively,” he said. “We want that on-demand component because the first and last miles are harder to cover.”
Van Hentenryck’s project will use U-M, Ann Arbor, and Detroit—including the Ann Arbor-Detroit corridor, which is loaded with commuters—as testing grounds for these innovative models, and will include collaboration with the U-M Parking and Transportation Services Department for real-time data collection on driver behavior. In Detroit, Van Hentenryck’s team will look for neighborhoods that would benefit from, or even be rejuvenated by, an on-demand system that increases mobility and keeps transit costs low.
“We want to understand people first, so we’ll mine the huge amount of data available in the Ann Arbor-Detroit region to understand what people are doing, when and where they’re going, and how,” he added. “People are very predictable, in a sense. We can build models at the aggregate level and use our infrastructure more effectively.”
Van Hentenryck said one of the biggest challenges will be simply parsing the massive amount of raw data. He gave the example of data from the Ann Arbor bus system; researchers will be able to access boarding data but they won’t know when a rider gets off the bus.
“We have to connect the datasets,” he said. “We have lots of information about one part, but not the whole.” He expects to find out why the transportation system is currently organized the way it is: “There may be a reason beyond simplicity.”
Once the two projects are able to analyze the data to the point that researchers can offer recommendations, the final hurdle will be to convince people to alter their behavior. In Ann Arbor, the public transit system is already at 75 percent capacity. In Detroit, it’s only at about 25 percent capacity, which is why Van Hentenryck is so keen to test new models there.
“The next couple of years will be very exciting,” he said. “Autonomous vehicles will benefit public transportation models immensely. It’s the convergence of a lot of ideas—data science, but transportation is also changing completely. In cities without a history of good public transportation, can they benefit? We want to see how people respond.”