*This was a HYBRID Event with in-person attendance in Levine 307 and Virtual attendance via Zoom
Frictional contact is the core underlying behavior of robot locomotion and manipulation, and its nearly-discontinuous dynamics make planning and control challenging even when an accurate model of the robot is available. In this talk, I will first present empirical evidence that learning an accurate model in the first place can be confounded by contact, as modern deep learning approaches are not designed to capture this non-smoothness. Second, I will discuss ContactNets, our approach which circumvents this conflict via a smooth, implicit encoding of discontinuity as signed distance functions and contact-frame Jacobians.