Back in September, we brought you news that UM-Dearborn scored a National Science Foundation grant to purchase a fully autonomous research vehicle, which Assistant Professor Jaerock Kwon expects will arrive on campus later this year. When it does, it’s going to be a huge boost for several faculty doing work in the autonomy space, primarily because it will provide an opportunity to do real-world testing of ideas they’ve been working on in simulation for years. While working with a full-size, fully autonomous vehicle is the ideal scenario for researchers, it turns out it’s not the only avenue for real-world testing. For the past couple years, graduate students in a lab led by Automotive and Mobility Systems Engineering (AMSE) Program Director Taehyun Shim have been testing autonomous features like lane keeping, parallel parking and emergency braking using souped-up toy cars that are about one-tenth the scale of an actual vehicle. Because they’re equipped with the similar camera-, radar- and lidar-based sensor systems, Shim says the scale cars actually allow for testing of similar types of perception algorithms used in real AVs. But the smaller vehicles have one huge advantage: They cost a fraction of the $244,610 that was recently granted to UM-Dearborn to purchase a real AV.
In fact, the scale car lab got its start on a shoestring budget of about $3,000. A few years ago, then-master’s student Arun Joshi was working with Shim on algorithms for autonomous lane keeping, object detection and emergency braking systems, and his work looked promising in simulation. But Joshi knew that the real test would be to see how his programming performed in an actual vehicle. When you ask algorithms to take the leap from simulation to the real world, Shim and Joshi say all kinds of issues can arise. The functioning of the various electrical components can introduce “noise” that can muddle signals coming from the sensors. You might not have enough onboard computing power to process all the information coming from the sensors in real time. Even mounting all the sensors to the car chassis in a way that the vehicle can still handle properly can be tricky. Joshi knew he wouldn’t have access to a real AV, but when he started doing some searching online, he ran across a possible alternative. It turned out there was a whole community of researchers building autonomous toy cars for racing as a way to advance autonomous technologies. He contacted some of the researchers at UC-Berkeley who were working in this space, who generously shared their plans. Shim loved the idea and helped Joshi score a small grant from the university, so Joshi could make the autonomous scale cars the subject of his master’s capstone project.
Joshi built the first of the autonomous scale cars in 2019, which he used to fine-tune his lane keeping and object detection algorithms. As expected, having an opportunity to do real-world testing was kind of a game changer. “In simulation, you can get it very close to perfect. But then when you’re dealing with the actual sensors functioning in a real environment, you run into issues that force you to account for all these things the simulation doesn’t include. You want the project to be successful, so you naturally begin to work on how to address those pitfalls, and you end up with a much better system. I mean, we didn't get it perfect, but it was much better than if we had only tested it in simulation.”
In subsequent years, other students have built on Joshi’s work. One year, a CECS undergraduate senior design team designed and built a four-meter track for the autonomous scale cars. Another master’s student, Pavan Anandareddy, used the scale cars to develop and test automated parallel parking algorithms. And one of Shim’s current graduate students, Syed Adil Ahmed, is working on lane keeping and adaptive cruise control systems. Shim says they now hope to use the scale car lab in the Automotive and Mobility Systems Engineering master’s program, which was updated this year to emphasize new technologies like intelligent vehicle systems and electrification.
Shim also says this kind of project-based work has huge benefits for students’ education and career prospects. “When you work on a project like this, I think the main benefit is you end up with this more complete understanding of the vehicle,” Shim says. “They have to work with the hardware, they have to do the programming, they have to build the vehicles and make all the systems work together. I think more and more, that’s the kind of holistic thinking that industry is looking for, because you’re able to troubleshoot more complex challenges when you have this more complete understanding.”
Joshi and Anandareddy can both speak to that. After graduation, Anandareddy landed a job at the electric truck startup Rivian. Joshi is now working in the autonomous vehicle division at Ford. “I think the way the auto industry is today, you have to have knowledge of a multitude of things,” Joshi says. “It’s all moving toward electric cars and autonomous systems, so you can’t confine yourself to knowledge of a particular thing. You sort of have to understand the whole picture. And now that I’m in the industry, I can see it’s not all that different from what we got to do with the scale cars. So I’m very thankful to have had the opportunity to do that.”
###
Are you interested in learning more about autonomous vehicles at UM-Dearborn? Check out the newly refreshed Automotive and Mobility Systems Engineering master’s program, which has new concentrations in electrification and intelligent vehicle systems. Story by Lou Blouin
