You’ve been working on automated and connected vehicles for a long time. What issues do you see ahead of us in these fields that need to be addressed?
Connected and automated vehicle technologies are not new and the promise that they can improve the safety and efficiency of the transport system has been known for the last 20 to 30 years. But there are barriers to using these technologies. Some of these barriers are political, regulatory, or legal, but some are truly technical and scientific. I’m not trying to remove any policy barriers. Our goal is to resolve some of these scientific issues so these technologies can be deployed. But to some extent our work also helps the regulatory body and the government to design better policies so that they can facilitate the deployment and development of these technologies.
We have designed the Mcity Autonomous Safety Assessment Program in the hope of providing the government a blueprint for ensuring the safety of autonomous vehicles before they are deployed. That could help both autonomous vehicle developers and the public to gain confidence. Right now, there’s a lot of potential but also a lack of trust in autonomous vehicle safety. When a Cruise robotaxi had a collision in San Francisco, that not only impacted the company but also dramatically affected the field itself. To realize the promise of automated and connected vehicles, technology companies like Commsignia and Cruise need other parts of society to support their efforts, such as government agencies from the perspective of policy makers.
What direction do you see that might work?
Recent reports show how important safety testing is for automated vehicles, which accounts for half of the development cost. There are three major approaches: testing in public is the most broadly used approach, but there are two issues with that. If the vehicle isn’t safe enough, crashes can occur. The other is that safety-critical situations are still rare. For a large proportion of the miles driven, vehicles do not encounter any challenging situations. The second approach is to try to solve these issues in simulations. There are still problems with this. Either it’s not realistic enough, or it’s inefficient and wastes a lot of computing hours.
We designed and developed a naturalistic driving environment simulator based on a machine learning model to test the safety performance of vehicles. For the simulator, we have trained the background agents so that we know at what time, which agent will perform which maneuver, and with what probability. Basically performs as a bad driver to challenge the vehicles. We did this in a scientific way to guarantee the unbiased nature of the simulation and testing. We then combine the simulator with the physical Mcity test facility to challenge real vehicles with these virtual agents and see how they respond.
Will you also test connectivity solutions with agents? How does connected vehicle technology fit into this?
Mcity
The University of Michigan created Mcity to bring together the diverse expertise and resources required to realize the potential of emerging mobility technologies, and their commercial and economic viability. Mcity is a public-private partnership and operates the world’s first purpose-built proving ground for testing the performance and safety of connected and automated vehicles and technologies under controlled and realistic conditions.
UMTRI
The University of Michigan Transportation Research Institute (UMTRI) was founded in 1965 as a global leader in transportation research and a partner of choice for industry leaders, foundations, and government agencies. UMTRI’s multidisciplinary research includes short and long-term projects in areas involving social and behavioral analyses, accident data collection, traffic safety analysis, and standards development and testing, as well as the deployment and evaluation of new safety and mobility technologies.
We know current automated vehicle technology is not a consumer product yet, mostly because of the safety performance. There are multiple technologies available to improve safety, and one of those is connectivity. Either by improved sensing capabilities or by infrastructure that supports automated vehicles.
From the vehicles’ perspective, connectivity serves as sort of a sixth sense. Onboard sensors have limitations in range and in line-of-sight. With connected vehicle technology, drivers can receive additional information about their surrounding driving environment, because other vehicles broadcast their location, speed and status as well. The infrastructure can also send information to the vehicles and use the information coming from connected vehicles to manage traffic flow as well as improve safety and efficiency. It can be mutually beneficial.
The other benefit is related to vulnerable road users, such as bicyclists and pedestrians. Oftentimes there’s miscommunication between vulnerable road users and vehicle drivers that could lead to crashes. V2X has the potential to improve communication between them.
How do you plan to use your results?
Our vision is to build this testing ground for different sectors throughout the ecosystem so they can come and work collaboratively to test out their systems. There’s an intersection in the Mcity test facility that is equipped with multiple sensors from a variety of vendors. We have multiple connected vehicle roadside units. We also have vehicles with onboard sensors. It’s a controlled and closed environment where vendors and OEMs can safely evaluate their systems and identify issues. Whether their roadside perception sensors produces accurate results. Whether a V2X message can be sent real-time or with a delay. Whether the messaging had any packet loss. We are able to help them identify issues.
Our other goal is to deploy these technologies in the real world. For example, the U-M Transportation Research Institute (UMTRI) has been awarded grants from Federal Highway and USDOT to deploy this technology through the Ann Arbor Connected Environment and Smart Intersections Project. Autonomous vehicle developers can use real-world intersections to leverage information from sensor data sharing messages and incorporate that with their motion planning and path planning algorithms and see how these will help.
What’s the benefit of this program in your work as a professor?
Actually, education and workforce development is our number one goal. We do research that pushes the boundaries, but more importantly, we train the next-generation workforce. We have programs like the Perot Jain TechLab at Mcity program in which we invite startup companies in the field of connected and automated vehicles to work with groups of five to six students on specific projects. A lot of these are technology development projects that are solving specific problems. It has been very successful with several students being hired by these companies following graduation.
