Computer Vision News - December 2022

5 Chelsea Finn I then dabbled a little bit in computational biology as well as other areas like robotics and computer vision and ended up really liking the problem of studying intelligence and robotics, especially robotics, because it's very much going all the way to a very tangible and real system. You are asking questions and from time to time you are also getting answers… Yeah, definitely. What answer has satisfied you the most? That's a really hard question. I think that there's been lots of work that's been quite satisfying. I feel like any research project provides some answers but also opens a lot more questions. Some of the things that I've been satisfied with are when we actually see robots doing things in the real world. In some of the early projects that I did, we were able to train robots to do tasks that are pretty basic for people but are fairly complex for robots, like being able to screw a cap onto a bottle or being able to use a patch flow to lift an object. Seeing the robot do those things was really fascinating and interesting. But at the same time, it opened up questions like, well, it can do that sort of skill in one environment. Now, can we have robots execute that sort of behavior in a variety of environments with a variety of different objects? From there, I've been kind of satisfied by results where we've been able to allow robots to do things with novel objects. There was one paper a few years ago where we would give a robot some objects, and we would tell it a goal. The goal is to move a couple of pieces of trash over to the side of a bin, and it would figure out that it can pick up this object that it has never seen before and use that to sweep the objects to the side of the bin without having being told that it More than 100 inspiring interviews with successful Women in Computer Vision in our archive Chelsea, tell us about your work! I do research in machine learning and robotics. I'm really fascinated by this question of how we might allow agents, including robots, to develop broadly intelligent behavior in the real world. I think that machine learning and interaction with the world is a key component of doing that. My work has looked at questions that I think are important in trying to solve that problem, including the ability to quickly learn new things by leveraging previous experience rather than learning from scratch. Also, the ability to generalize broadly by using a wide variety of data and by training models to be more robust. Also, it has involved applying machine learning to real robots and seeing if they can learn manipulation skills in the real world. That entails both perception and action for robots to be able to see and to be able to act in the world. When did you discover this passion? It's something that kind of happens over time. I don't think there's any one moment in which you're like, “ Oh, this is the one thing that I love doing! ” I've always enjoyed solving puzzles and trying to figure out answers to questions. I think that this question of developing intelligent robots is a really fascinating one to work on. I did a little bit of robotics and computer science in middle school. I enjoyed working on it, but I didn't really think of it as something that I would dedicate my career to at that point. And then later on, when I was in college, it kind of became clear to me that computer sciencewould open a lot of doors toward solving really interesting problems.