August 09, 2023

Artificial intelligence (AI) makes all sorts of decisions for us. Netflix recommends our next movie. Amazon suggests products based on what’s in our shopping cart. Facebook determines the content that shows up in our feeds.

While we have a basic idea of how these algorithms work, most of us don’t need machines to further explain why they are making those conclusions.

But when it comes to entrusting our safety to self-driving cars, we need to know what triggers a computerized decision such as stopping the car or accelerating.

That’s where Mushuang Liu comes in. An assistant professor in mechanical and aerospace engineering. Liu is working to improve the algorithms behind vehicles that can drive themselves.

“We need the algorithm to always ensure that self-driving cars are safe, reliable and trustworthy,” she said. “But that’s challenging because it requires knowing the intentions of other vehicles and drivers.”

To better calculate those unknowns, Liu is using game theory, a branch of mathematics that takes into consideration decision-making situations where outcomes depend on choices made by others. Specifically, she’s developing a receding horizon potential game approach, which not only considers the needs of pedestrians, cyclists and other road participants to model their behaviors, but also enables extended prediction horizon for the self-driving cars. By integrating the merits of model predictive control and game theory, the algorithms can anticipate everyone’s varying goals and behaviors and respond accordingly.

Liu has funding from both Ford and MU to focus on creating a methodical approach to develop advanced control systems for automated driving. The goals are to ensure safety and enhance passenger comfort. She and collaborators use statistical and comparative simulation to test methods to determine how different decisions work in real-world situations, such as crossing intersections or changing lanes.

At Mizzou, Liu is hoping to team up with others working in various application areas.

“The developed algorithms and theories are not constrained by applications,” she said. “Effective optimizations can benefit heterogeneous applications, ranging from robotics to smart material, smart manufacturing, and smart health care. I am open to all possibilities!”

Liu also welcomes student researchers, especially those with a solid mathematical background.

Anyone interested can contact her at ml529@missouri.edu.