A Compositional Object-Based Approach to Learning Physical Dynamics
arXiv:1612.00341
Abstract
We present the Neural Physics Engine (NPE), a framework for learning simulators of intuitive physics that naturally generalize across variable object count and different scene configurations. We propose a factorization of a physical scene into composable object-based representations and a neural network architecture whose compositional structure factorizes object dynamics into pairwise interactions. Like a symbolic physics engine, the NPE is endowed with generic notions of objects and their interactions; realized as a neural network, it can be trained via stochastic gradient descent to adapt to specific object properties and dynamics of different worlds. We evaluate the efficacy of our approach on simple rigid body dynamics in two-dimensional worlds. By comparing to less structured architectures, we show that the NPE's compositional representation of the structure in physical interactions improves its ability to predict movement, generalize across variable object count and different scene configurations, and infer latent properties of objects such as mass.
Published as a conference paper for ICLR 2017. 15 pages, 6 figures
Cited by in corpus (12)
- A Disentangled Recognition and Nonlinear Dynamics Model for Unsupervised Learning
- Explainability Techniques for Graph Convolutional Networks
- Learning A Physical Long-term Predictor
- Learning Robotic Manipulation of Granular Media
- Bounce and Learn: Modeling Scene Dynamics with Real-World Bounces
- Programmable Agents
- Unsupervised Intuitive Physics from Past Experiences
- Efficient Cloth Simulation using Miniature Cloth and Upscaling Deep Neural Networks
- Structured agents for physical construction
- Neural Consciousness Flow
- Adding Intuitive Physics to Neural-Symbolic Capsules Using Interaction Networks
- cvpaper.challenge in 2016: Futuristic Computer Vision through 1,600 Papers Survey