SEAS faculty safely explore interactions between autonomous vehicles and humans

By Elizabeth Egan 

Virtual simulation offers a solution, providing the ability to place human participants in artificial environments with simulated CAVs to examine human performance behavior without simultaneously endangering the safety of a human participant who is having to drive, walk or cycle around a test track alongside an unproven self-driven vehicle. However, virtual simulation comes with limitations as to how accurately it reflects the behaviors of real human participants.

In a collaborative effort between the Department of Civil, Structural and Environmental Engineering (CSEE) and the Department of Mechanical and Aerospace Engineering (MAE), researchers from the School of Engineering and Applied Sciences are working to overcome these roadblocks in testing CAV safety. The research team consists of Chaozhe He, MAE assistant professor, Austin Angulo, CSEE assistant professor, Irina Benedyk, CSEE assistant professor, and Kevin Hulme, program manager for the Stephen Still Institute for Sustainable Transportation and Logistics (ISTL). They have gained access to an innovative facility that will allow them to integrate human participants within a virtual environment and develop mixed reality testing capabilities using the University at Buffalo’s brand new Transportation Research and Visualization Laboratory (TRAVL).

The team requested and received approval to access Mcity 2.0, a world-class testing facility based out of the University of Michigan. Their proposal was one of three selected.

Mcity is a 32-acre testing site at the University of Michigan that provides a controlled environment to test automated vehicles before they are deployed on public streets. Mcity 2.0 was created with a $5.1 million grant from the National Science Foundation to develop a digital infrastructure overlay to the Mcity facility. Mcity 2.0 is a cloud-based, augmented-reality CAV testbed that could be made available worldwide. The facility is equipped with a networking communications platform that offers a way to communicate between the digital twin environment and the test track.

“What this means, at the end of the day, is that we can have multiple people in simulators here at UB, controlling real vehicles and dummies on the test track at Mcity, interacting with real CAVs on the test track, all within a virtual simulation,” said Angulo, who is also the director of the TRAVL Lab. “It’s a bit mind bending, but incredible!”

The UB researchers will be using the facility to conduct multi-agent testing and verification of CAV operations with human drivers and vulnerable road users, including pedestrians and people on bikes or scooters.

After integrating the multi-agent physical presence at the TRAVL Lab virtually within the digital twin of the Mcity facility, the team will seek to enable the remote presence of vehicles and vulnerable road users within the virtual simulators at Mcity. Taking advantage of the physical equipment in the TRAVL Lab along with the digital components at Mcity will allow the team to  jointly test interactions with real vehicles on the physical test track, extending reality testing capability at Mcity.

“Our work would enhance current TRAVL capabilities and strengthen UB's position in CAV research,” said He, the project’s principal investigator. “The investigation is the first of its kind in the world. Once it succeeds, we will be in a stronger position to pursue external funding and conduct impactful research.”

One aspect of the study that Angulo noted involves testing the relationship between CAVs and bicycles when a CAV passes in a shared lane.

“This may sound like a rather mundane scenario, but it is a rather dangerous one,” noted Angulo. “There is very little data that exists for evaluating CAV safety performance and no data exists for understanding bicyclist experiences or preferences.”

The team will use the remote capabilities to run scenarios between a CAV at the Mcity test track and a bicyclist dummy. They hope to answer important questions such as: at what speed should the CAV pass the bicyclist, how well does the CAV react to sudden changes in the bicyclist’s behavior, does the presence of a parked car on the curb impact how much space the CAV should leave between itself and the bicycle, and more.

The testing can be taken a step further to evaluate the bicyclist’s lane position, speed, breaking, eye tracking and more to understand how the person responds to the passing CAV and what behaviors they might see as unsafe, stressful or distracting.

“These are all very relevant questions that need answering so that we can better design CAV safety behavior and operations,” said Angulo. “This data will provide invaluable insight into the experiences of vulnerable road users and how to improve CAV behavior to enable all modalities on the road.”