Predicting a Mouse's Movement

Getting a mouse to eat a piece of cheese is not an inherently difficult task.

But getting a mouse to eat a piece of cheese while its head is fixed in place so its natural movements can be recorded? There is nothing natural about that.

That was the reason the Shepherd Lab at Northwestern's Feinberg School of Medicine partnered with three students in Northwestern Engineering's Master of Science in Artificial Intelligence (MSAI) program.

Namitha George (MSAI '26), Amitha Gowda (MSAI '26), and Jianchen Hong (MSAI '26) worked with the Shepherd Lab for their Practicum in Intelligent Systems course. The practicum is meant to help students apply lessons learned in the classroom as they design and develop solutions to an open-ended AI problem.

“I was excited by how interdisciplinary the project was,” Namitha said. “It combined computer vision and machine learning to solve a real problem for the lab that would help further their research.”

That research is focused on increasing understanding of how the brain controls dextrous, coordinated movement. To do that, researchers examine how different areas of the motor cortex communicate with each other to accomplish complex motions. A mouse moving toward a piece of cheese and scooping it up to eat might appear simple, but it actually is an intricate series of mini-movements that make up the larger action.

The challenge the MSAI students faced related to video motion-capture technology. The lab previously relied on a series of cameras recording a static mouse’s body-part movements, but by locking the mouse’s head in place to capture its motion, the movement recorded was hardly natural.

“Most things you do almost without thinking involve much more complicated behaviors,” said John Barrett, a research assistant professor at Feinberg and adviser for the students' project. “They involve different body parts working together and it's hard to get the mouse to do those sort of really interesting behaviors if you hold its head in place.”

The students were asked to create and calibrate a system involving an additional camera that accurately tracks the mouse’s movement. They used large data sets on mouse motion to create an algorithm that allowed the camera to predict where the tiny rodent was headed before it went there, meaning it could track the motion in real time without lag.

“I liked that there wasn't a predefined solution,” Namitha said. “We had to research different approaches, test ideas, and adapt as we learned more about the challenges. It was rewarding to know that this project could help researchers better understand how the brain controls movement.”

John appreciated the students' dedication to the project.

“They made a lot of progress on figuring out what the issues are with getting all these data streams working together,” he said. “We quickly realized that we weren't giving them as much data as they needed, and just knowing that is helpful and should hopefully set us up for the next iteration of the project.”

John said the lab will continue to work with MSAI students moving forward. The students, meanwhile, said they appreciated the practical nature of the practicum project.

It was an experience they said will be beneficial as they finish MSAI and pursue full-time job opportunities.

“It’s important to apply the skills we learn in the classroom to real problems clients face,” Namitha said. “Many of us in the MSAI program will be working in industry after graduation, and things like requirements gathering, understanding scope of work, and asking questions are all vital skills in a project. Practicum is a great opportunity to develop those skills.”