Get our free extension to see links to code for papers anywhere online!

Chrome logo  Add to Chrome

Firefox logo Add to Firefox

Learning Contact Dynamics using Physically Structured Neural Networks

Add code

Feb 22, 2021
Andreas Hochlehnert, Alexander Terenin, Steind贸r S忙mundsson, Marc Peter Deisenroth

Share this with someone who'll enjoy it:

Learning physically structured representations of dynamical systems that include contact between different objects is an important problem for learning-based approaches in robotics. Black-box neural networks can learn to approximately represent discontinuous dynamics, but they typically require large quantities of data and often suffer from pathological behaviour when forecasting for longer time horizons. In this work, we use connections between deep neural networks and differential equations to design a family of deep network architectures for representing contact dynamics between objects. We show that these networks can learn discontinuous contact events in a data-efficient manner from noisy observations in settings that are traditionally difficult for black-box approaches and recent physics inspired neural networks. Our results indicate that an idealised form of touch feedback -- which is heavily relied upon by biological systems -- is a key component of making this learning problem tractable. Together with the inductive biases introduced through the network architectures, our techniques enable accurate learning of contact dynamics from observations.

* Artificial Intelligence and Statistics, 2021 

   Access Paper Source

Share this with someone who'll enjoy it: