Visual Grounding of Learned Physical Models

Part of Proceedings of the International Conference on Machine Learning 1 pre-proceedings (ICML 2020)

Bibtex »Metadata »Paper »

Bibtek download is not availble in the pre-proceeding


Authors

Yunzhu Li, Toru Lin, Kexin Yi, Daniel Bear, Daniel Yamins, Jiajun Wu, Josh Tenenbaum, Antonio Torralba

Abstract

<p>Humans can intuitively recognize objects’ physical properties and predict their future motion, even when the objects are engaged in complicated interactions with each other. The ability to perform physical reasoning and adapt to new environments, while intrinsic to humans, remains challenging to state-of-the-art computational models. In this work, we present a neural model that simultaneously reasons about physics and make future predictions based on visual and dynamics priors. The visual prior predicts a particle-based representation of the system from visual observations. An inference module operates on those particles, predicting and refining estimates of particle locations, object states, and physical parameters, subject to the constraints imposed by the dynamics prior, which we refer to as visual grounding. We demonstrate the effectiveness of our method in environments involving rigid objects, deformable materials, and fluids. Experiments show that our model can infer the physical properties within a few observations, which allows the model to quickly adapt to unseen scenarios and make accurate predictions into the future.</p>