Researchers from the Mcity Test Facility at the University of Michigan are using virtual reality to more safely and efficiently test driverless vehicles.
The Mcity Test Facility is located on North Campus and features over 16 acres of roads and traffic infrastructure. The facility allows researchers to test automated vehicles and other technologies in simulated urban and suburban driving environments.
With a virtual world, similar to the simulated environment of a video game, researchers can test real driverless vehicles in an almost infinite number of traffic scenarios that would be difficult and costly to test in real world environments. In the augmented reality testing environment, real autonomous vehicles interact with computer-simulated vehicles.
The testing method of using augmented reality was outlined in an MCity white paper released last week. Henry Liu, a research professor at the University Transportation Research Institute, developed this testing technology with UMTRI assistant research scientist Yiheng Feng.
Currently, Liu and Feng have a patent pending for the testing technology.
Liu explained while augmented reality is not exactly the same as a real world environment, testing autonomous vehicles in a virtual environment eliminates many of the restrictions of testing in the real world.
“Testing driverless vehicles in an augmented reality environment is not going to be as ‘accurate’ as testing in the ‘real world,’” Liu wrote in an email interview. “However, testing in the real world has a number of limitations that are difficult to overcome.”
Liu cited safety as a top limitation and concern when testing autonomous vehicles in real world environments. Testing in a virtual environment is safer, he said, because there have been many driverless vehicle accidents when testing in the real world. According to Liu and Feng’s publication, there have been 26 crashes while testing autonomous vehicles at their site in California from 2014 to 2017.
Liu explained augmented reality also makes testing autonomous vehicles more efficient because researchers can simulate rare traffic scenarios that may be difficult to find in the real world.
“Most of the testing in the real world is not very useful, as the testing environment is repetitive and it is not challenging to autonomous vehicles,” Liu wrote. “What is important is to find ‘corner cases’ that can challenge autonomous vehicles, but ‘corner cases’ are rare events so it is not easy to find in the real world.”
Mcity Director Huei Peng Huei Peng supported the use of augmented reality in the testing of University driverless vehicles.
“Since virtual vehicles are used, instead of real vehicles, we can simulate risky situations without the real safety risk, we do not need to waste fuel and we can run continuously,” Peng said. “So it is safer, cheaper and faster compared with using real test vehicles.”
Engineering senior Michelle Kearney has been a research assistant at UMTRI for her entire undergraduate career. At Mcity, she has been involved with several studies on passenger behavior and responses to autonomous vehicles.
“Our recent tests at MCity have focused on the safety of individuals who are actively engaged in a task other than driving (i.e. working on a laptop, reading a book, etc.) and are subjected to an abrupt automatic braking or lane change by the vehicle,” Kearney wrote in an email interview. “In addition to safety considerations, the study also provided great insight into consumer response to unanticipated abrupt vehicle movements.”
Kearney said using augmented reality for testing autonomous vehicles will help advance research of consumer safety.
“Simulators can be very important tools in autonomous vehicle research, as they allow researchers to gather key information about consumer preferences and behaviors with regards to autonomous vehicles without requiring individuals to trust the actual vehicle for a test drive,” Kearney wrote. “It is going to be a slow process to convince the public that these vehicles are safe and reliable, so using virtual reality simulation is a good first step.”