Augmenting Physical Simulators with Stochastic Neural Networks: Case Study of Planar Pushing and Bouncing
© 2018 IEEE. An efficient, generalizable physical simulator with universal uncertainty estimates has wide applications in robot state estimation, planning, and control. In this paper, we build such a simulator for two scenarios, planar pushing and ball bouncing, by augmenting an analytical rigid-bod...
| Main Authors: | , , , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2021
|
| Online Access: | https://hdl.handle.net/1721.1/137711 |
| _version_ | 1811088729929940992 |
|---|---|
| author | Ajay, Anurag Wu, Jiajun Fazeli, Nima Bauza, Maria Kaelbling, Leslie P. Tenenbaum, Joshua B. Rodriguez, Alberto |
| author2 | Massachusetts Institute of Technology. Department of Mechanical Engineering |
| author_facet | Massachusetts Institute of Technology. Department of Mechanical Engineering Ajay, Anurag Wu, Jiajun Fazeli, Nima Bauza, Maria Kaelbling, Leslie P. Tenenbaum, Joshua B. Rodriguez, Alberto |
| author_sort | Ajay, Anurag |
| collection | MIT |
| description | © 2018 IEEE. An efficient, generalizable physical simulator with universal uncertainty estimates has wide applications in robot state estimation, planning, and control. In this paper, we build such a simulator for two scenarios, planar pushing and ball bouncing, by augmenting an analytical rigid-body simulator with a neural network that learns to model uncertainty as residuals. Combining symbolic, deterministic simulators with learnable, stochastic neural nets provides us with expressiveness, efficiency, and generalizability simultaneously. Our model outperforms both purely analytical and purely learned simulators consistently on real, standard benchmarks. Compared with methods that model uncertainty using Gaussian processes, our model runs much faster, generalizes better to new object shapes, and is able to characterize the complex distribution of object trajectories. |
| first_indexed | 2024-09-23T14:06:37Z |
| format | Article |
| id | mit-1721.1/137711 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T14:06:37Z |
| publishDate | 2021 |
| publisher | IEEE |
| record_format | dspace |
| spelling | mit-1721.1/1377112023-02-03T21:01:05Z Augmenting Physical Simulators with Stochastic Neural Networks: Case Study of Planar Pushing and Bouncing Ajay, Anurag Wu, Jiajun Fazeli, Nima Bauza, Maria Kaelbling, Leslie P. Tenenbaum, Joshua B. Rodriguez, Alberto Massachusetts Institute of Technology. Department of Mechanical Engineering Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory © 2018 IEEE. An efficient, generalizable physical simulator with universal uncertainty estimates has wide applications in robot state estimation, planning, and control. In this paper, we build such a simulator for two scenarios, planar pushing and ball bouncing, by augmenting an analytical rigid-body simulator with a neural network that learns to model uncertainty as residuals. Combining symbolic, deterministic simulators with learnable, stochastic neural nets provides us with expressiveness, efficiency, and generalizability simultaneously. Our model outperforms both purely analytical and purely learned simulators consistently on real, standard benchmarks. Compared with methods that model uncertainty using Gaussian processes, our model runs much faster, generalizes better to new object shapes, and is able to characterize the complex distribution of object trajectories. 2021-11-08T16:58:37Z 2021-11-08T16:58:37Z 2018-10 2019-06-04T15:47:05Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/137711 Ajay, Anurag, Wu, Jiajun, Fazeli, Nima, Bauza, Maria, Kaelbling, Leslie P. et al. 2018. "Augmenting Physical Simulators with Stochastic Neural Networks: Case Study of Planar Pushing and Bouncing." en 10.1109/iros.2018.8593995 Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf IEEE MIT web domain |
| spellingShingle | Ajay, Anurag Wu, Jiajun Fazeli, Nima Bauza, Maria Kaelbling, Leslie P. Tenenbaum, Joshua B. Rodriguez, Alberto Augmenting Physical Simulators with Stochastic Neural Networks: Case Study of Planar Pushing and Bouncing |
| title | Augmenting Physical Simulators with Stochastic Neural Networks: Case Study of Planar Pushing and Bouncing |
| title_full | Augmenting Physical Simulators with Stochastic Neural Networks: Case Study of Planar Pushing and Bouncing |
| title_fullStr | Augmenting Physical Simulators with Stochastic Neural Networks: Case Study of Planar Pushing and Bouncing |
| title_full_unstemmed | Augmenting Physical Simulators with Stochastic Neural Networks: Case Study of Planar Pushing and Bouncing |
| title_short | Augmenting Physical Simulators with Stochastic Neural Networks: Case Study of Planar Pushing and Bouncing |
| title_sort | augmenting physical simulators with stochastic neural networks case study of planar pushing and bouncing |
| url | https://hdl.handle.net/1721.1/137711 |
| work_keys_str_mv | AT ajayanurag augmentingphysicalsimulatorswithstochasticneuralnetworkscasestudyofplanarpushingandbouncing AT wujiajun augmentingphysicalsimulatorswithstochasticneuralnetworkscasestudyofplanarpushingandbouncing AT fazelinima augmentingphysicalsimulatorswithstochasticneuralnetworkscasestudyofplanarpushingandbouncing AT bauzamaria augmentingphysicalsimulatorswithstochasticneuralnetworkscasestudyofplanarpushingandbouncing AT kaelblinglesliep augmentingphysicalsimulatorswithstochasticneuralnetworkscasestudyofplanarpushingandbouncing AT tenenbaumjoshuab augmentingphysicalsimulatorswithstochasticneuralnetworkscasestudyofplanarpushingandbouncing AT rodriguezalberto augmentingphysicalsimulatorswithstochasticneuralnetworkscasestudyofplanarpushingandbouncing |