David Fouhey in his office.

Fouhey, in his office at the College of Engineering between meetings with students.

David Fouhey wants your computer to be as smart as your cat.

He’s working pretty hard to make it happen, and he’s got a big group of U-M students on board in his Computer Vision class in the College of Engineering and some great research collaborations to help make that happen.

In all reality, Fouhey wants your computer to learn how to learn things. He wants it to know how to deal with unknown information: just because you can’t see all four legs of the table, that doesn’t mean they aren’t there. How do you navigate around a kitchen island when you can’t see the other side of it? Or what’s inside a drawer and how does it open?

Computer vision has some grasp on 2D spatial relationships, but real world scenarios are more problematic. Understanding the next level – 3D spaces – is crucial to AI applications in robotics, vision, planning and reasoning. While this is completely obvious to you or even to a cat, it’s totally missing to computers.

“We have this common sense understanding of how the world works and much of that is based on space and how things are organized, how physical objects work together. The goal is for computers to have this level of understanding and learn much of this automatically. Computers need to have common sense about the physical world.”

Welcome to the team

Fouhey is one of the newest faculty members at the College of Engineering. After a Ph.D. at Carnegie Mellon University and postdoc years at UC-Berkeley, he’s hit the ground running at the University of Michigan.

It can be a daunting task as new faculty to learn a new campus, prepare to teach, and secure funding for research. “My department is wonderful, they are extremely supportive. Senior faculty are keeping me on track and helping navigate the new job.” Students are keeping him busy in office hours, and his research is getting a lot of support.

Faculty are also opening doors for him in the research realm. Satinder Singh Baveja, a faculty member in computer science and engineering who holds the Toyota Professorship in Artificial Intelligence in the College of Engineering, realized quickly that Fouhey’s research would dovetail well with U-M’s partnership with Toyota Research Institute (TRI) and supported Fouhey’s grant application for a collaborative research project.

Fouhey’s research for “Building and Reasoning about Fully 3D Representations” received TRI funding and aligns the companies technological goals with Fouhey’s research objectives. He says, “the main goal of this research is to enable computers to learn to extract such a 3D representation from ordinary images and to connect this ability with tasks and settings that are relevant to autonomous systems, such as service robots indoors and autonomous vehicles.”

On the importance of this type of funding, Fouhey shares, “TRI funding has allowed me to quickly get off the ground and support my first PhD students. It is wonderful to have a sponsor like TRI that understands and is aligned with my long-term research goals.”

Umesh Patel at the Business Engagement Center shared, “We actively pursue partnerships like this where faculty see natural connections between existing research and company interests. Working together, we can advance discoveries further and faster and get them into technologies that improve the quality of lives.”

It’s ‘easy’ math

Fouhey says the math involved is fairly simple, but “if you put simple things together, you can create very complicated systems.”

Computer image of a living room

What your cat sees.

Computer vision of a room

What the computer model is able to envision, based on learning from an inordinate amount of image data.

What isn’t so simple is the inordinately large amount of that math that needs to happen and the data needed for it. Advancing to 3D machine learning is somewhat a function of data, and massive amounts of data need to be processed for computers to establish scenarios and learn to learn.

“I’m excited to work do this work with U-M and TRI. They have real and interesting problems to talk about and work on.  They have a lot of data they’re letting me play with, and that is amazing because it is really hard to get lots of good data like this. That’s one of the biggest issues in computer vision: you are limited by data you can use.”

Now Fouhey has spent a lot of time helping students push through the end of the semester. He is energized and engaged with them, and you get the feeling that this crop of students will be the ones that help really propel advances as they gain skills. Fresh ideas and approaches from new faculty and the students they teach help drive innovation.

With Fouhey leading them all, they just may be able to beat your cat.