End-to-end differentiable physics for learning and control
© 2018 Curran Associates Inc.All rights reserved. We present a differentiable physics engine that can be integrated as a module in deep neural networks for end-to-end learning. As a result, structured physics knowledge can be embedded into larger systems, allowing them, for example, to match observa...
| Main Authors: | , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
Curran Associates Inc
2020
|
| Online Access: | https://hdl.handle.net/1721.1/126615 |
| _version_ | 1811085434526105600 |
|---|---|
| author | Smith, Kevin A Allen, Kelsey Rebecca Tenenbaum, Joshua B |
| author2 | Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences |
| author_facet | Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences Smith, Kevin A Allen, Kelsey Rebecca Tenenbaum, Joshua B |
| author_sort | Smith, Kevin A |
| collection | MIT |
| description | © 2018 Curran Associates Inc.All rights reserved. We present a differentiable physics engine that can be integrated as a module in deep neural networks for end-to-end learning. As a result, structured physics knowledge can be embedded into larger systems, allowing them, for example, to match observations by performing precise simulations, while achieves high sample efficiency. Specifically, in this paper we demonstrate how to perform backpropagation analytically through a physical simulator defined via a linear complementarity problem. Unlike traditional finite difference methods, such gradients can be computed analytically, which allows for greater flexibility of the engine. Through experiments in diverse domains, we highlight the system's ability to learn physical parameters from data, efficiently match and simulate observed visual behavior, and readily enable control via gradient-based planning methods. Code for the engine and experiments is included with the paper. |
| first_indexed | 2024-09-23T13:09:32Z |
| format | Article |
| id | mit-1721.1/126615 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T13:09:32Z |
| publishDate | 2020 |
| publisher | Curran Associates Inc |
| record_format | dspace |
| spelling | mit-1721.1/1266152022-10-01T13:24:45Z End-to-end differentiable physics for learning and control Smith, Kevin A Allen, Kelsey Rebecca Tenenbaum, Joshua B Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences © 2018 Curran Associates Inc.All rights reserved. We present a differentiable physics engine that can be integrated as a module in deep neural networks for end-to-end learning. As a result, structured physics knowledge can be embedded into larger systems, allowing them, for example, to match observations by performing precise simulations, while achieves high sample efficiency. Specifically, in this paper we demonstrate how to perform backpropagation analytically through a physical simulator defined via a linear complementarity problem. Unlike traditional finite difference methods, such gradients can be computed analytically, which allows for greater flexibility of the engine. Through experiments in diverse domains, we highlight the system's ability to learn physical parameters from data, efficiently match and simulate observed visual behavior, and readily enable control via gradient-based planning methods. Code for the engine and experiments is included with the paper. 2020-08-17T15:06:34Z 2020-08-17T15:06:34Z 2018-12 2019-10-08T14:43:33Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/126615 Belbute-Peres, Filipe de A. et al. “End-to-end differentiable physics for learning and control.” Paper presented at the 32nd Conference on Neural Information Processing Systems (NeurIPS 2018), Montréal, Dec 3-8 2018, Curran Associates Inc © 2018 The Author(s) en https://papers.nips.cc/paper/7948-end-to-end-differentiable-physics-for-learning-and-control 32nd Conference on Neural Information Processing Systems (NeurIPS 2018) Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. application/pdf Curran Associates Inc Neural Information Processing Systems (NIPS) |
| spellingShingle | Smith, Kevin A Allen, Kelsey Rebecca Tenenbaum, Joshua B End-to-end differentiable physics for learning and control |
| title | End-to-end differentiable physics for learning and control |
| title_full | End-to-end differentiable physics for learning and control |
| title_fullStr | End-to-end differentiable physics for learning and control |
| title_full_unstemmed | End-to-end differentiable physics for learning and control |
| title_short | End-to-end differentiable physics for learning and control |
| title_sort | end to end differentiable physics for learning and control |
| url | https://hdl.handle.net/1721.1/126615 |
| work_keys_str_mv | AT smithkevina endtoenddifferentiablephysicsforlearningandcontrol AT allenkelseyrebecca endtoenddifferentiablephysicsforlearningandcontrol AT tenenbaumjoshuab endtoenddifferentiablephysicsforlearningandcontrol |