Anton Arapin

I'm a first-year PhD student at the University of Michigan, advised by Bernadette Bucher in the Mapping and Motion Lab. My research sits at the intersection of geometric generative models, computer vision, and robotics. I build world models for robotics — action-conditioned models that predict how a scene will change — with a focus on predicting and using uncertainty and preserving geometric structure as scenes evolve.

Before Michigan I was a software engineer at Google, working on a large-scale machine learning system for modeling and predicting user behavior, with a focus on the accuracy, quality, and efficiency of training and inference.

I've also worked with Junchen Jiang at the University of Chicago on neural video codecs, and with Kostas Daniilidis at the GRASP Lab on robot exploration and navigation. I also did research in projective geometry with Annalisa Crannell and reinforcement learning with Erin Talvitie.