CoRL 2020, Spotlight Talk 506: ContactNets: Learning Discontinuous Contact Dynamics with Smooth, Implicit Representations

CoRL 2020

CoRL 2020, Spotlight Talk 506: ContactNets: Learning Discontinuous Contact Dynamics with Smooth, Implicit Representations

Dec 16, 2020
|
27 views
|
Details
**ContactNets: Learning Discontinuous Contact Dynamics with Smooth, Implicit Representations** Samuel Pfrommer (University of Pennsylvania); Mathew Halm (University of Pennsylvania)*; Michael Posa (University of Pennsylvania) Publication: http://corlconf.github.io/paper_506/ **Abstract** Common methods for learning robot dynamics assume motion is continuous, causing unrealistic model predictions for systems undergoing discontinuous impact and stiction behavior. In this work, we resolve this conflict with a smooth, implicit encoding of the structure inherent to contact-induced discontinuities. Our method, ContactNets, learns parameterizations of inter-body signed distance and contact-frame Jacobians, a representation that is compatible with many simulation, control, and planning environments for robotics. We furthermore circumvent the need to differentiate through stiff or non-smooth dynamics with a novel loss function inspired by the principles of complementarity and maximum dissipation. Our method can predict realistic impact, non-penetration, and stiction when trained on 60 seconds of real-world data.

Comments
loading...