Learning to Represent Mechanics via Long-term Extrapolation and Interpolation

06/06/2017
by   Sebastien Ehrhardt, et al.
0

While the basic laws of Newtonian mechanics are well understood, explaining a physical scenario still requires manually modeling the problem with suitable equations and associated parameters. In order to adopt such models for artificial intelligence, researchers have handcrafted the relevant states, and then used neural networks to learn the state transitions using simulation runs as training data. Unfortunately, such approaches can be unsuitable for modeling complex real-world scenarios, where manually authoring relevant state spaces tend to be challenging. In this work, we investigate if neural networks can implicitly learn physical states of real-world mechanical processes only based on visual data, and thus enable long-term physical extrapolation. We develop a recurrent neural network architecture for this task and also characterize resultant uncertainties in the form of evolving variance estimates. We evaluate our setup to extrapolate motion of a rolling ball on bowl of varying shape and orientation using only images as input, and report competitive results with approaches that assume access to internal physics models and parameters.

READ FULL TEXT
research
12/22/2017

Taking Visual Motion Prediction To New Heightfields

While the basic laws of Newtonian mechanics are well understood, explain...
research
03/01/2017

Learning A Physical Long-term Predictor

Evolution has resulted in highly developed abilities in many natural int...
research
04/20/2021

Explainable artificial intelligence for mechanics: physics-informing neural networks for constitutive models

(Artificial) neural networks have become increasingly popular in mechani...
research
06/09/2020

Neural Physicist: Learning Physical Dynamics from Image Sequences

We present a novel architecture named Neural Physicist (NeurPhy) to lear...
research
01/20/2022

Symplectic Momentum Neural Networks – Using Discrete Variational Mechanics as a prior in Deep Learning

With deep learning being gaining attention from the research community f...
research
03/06/2023

A comparative study on different neural network architectures to model inelasticity

The mathematical formulation of constitutive models to describe the path...
research
09/30/2019

Development of a Game with a Purpose for Acquisition of Brain-Computer Interface Data

Brain-computer interfaces (BCIs) have the potential to significantly cha...

Please sign up or login with your details

Forgot password? Click here to reset