Combining physical simulators and object-based networks for control
Physics engines play an important role in robot planning and control; however, many real-world control problems involve complex contact dynamics that cannot be characterized analytically. Most physics engines therefore employ approximations that lead to a loss in precision. In this paper, we propose...
| Main Authors: | , , , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2020
|
| Online Access: | https://hdl.handle.net/1721.1/126674 |
| _version_ | 1811083578521419776 |
|---|---|
| author | Ajay, Anurag. Bauza Villalonga, Maria Wu, Jiajun Fazeli, Nima Tenenbaum, Joshua B Rodriguez Garcia, Alberto Kaelbling, Leslie P |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Ajay, Anurag. Bauza Villalonga, Maria Wu, Jiajun Fazeli, Nima Tenenbaum, Joshua B Rodriguez Garcia, Alberto Kaelbling, Leslie P |
| author_sort | Ajay, Anurag. |
| collection | MIT |
| description | Physics engines play an important role in robot planning and control; however, many real-world control problems involve complex contact dynamics that cannot be characterized analytically. Most physics engines therefore employ approximations that lead to a loss in precision. In this paper, we propose a hybrid dynamics model, simulator-augmented interaction networks (SAIN), combining a physics engine with an object-based neural network for dynamics modeling. Compared with existing models that are purely analytical or purely data-driven, our hybrid model captures the dynamics of interacting objects in a more accurate and data-efficient manner. Experiments both in simulation and on a real robot suggest that it also leads to better performance when used in complex control tasks. Finally, we show that our model generalizes to novel environments with varying object shapes and materials. |
| first_indexed | 2024-09-23T12:35:19Z |
| format | Article |
| id | mit-1721.1/126674 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T12:35:19Z |
| publishDate | 2020 |
| publisher | IEEE |
| record_format | dspace |
| spelling | mit-1721.1/1266742022-10-01T09:57:07Z Combining physical simulators and object-based networks for control Ajay, Anurag. Bauza Villalonga, Maria Wu, Jiajun Fazeli, Nima Tenenbaum, Joshua B Rodriguez Garcia, Alberto Kaelbling, Leslie P Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology. Department of Mechanical Engineering Physics engines play an important role in robot planning and control; however, many real-world control problems involve complex contact dynamics that cannot be characterized analytically. Most physics engines therefore employ approximations that lead to a loss in precision. In this paper, we propose a hybrid dynamics model, simulator-augmented interaction networks (SAIN), combining a physics engine with an object-based neural network for dynamics modeling. Compared with existing models that are purely analytical or purely data-driven, our hybrid model captures the dynamics of interacting objects in a more accurate and data-efficient manner. Experiments both in simulation and on a real robot suggest that it also leads to better performance when used in complex control tasks. Finally, we show that our model generalizes to novel environments with varying object shapes and materials. NSF (nos. 1420316, 1523767, and 1723381) AFOSR (Grant FA9550- 17-1-0165) ONR MURI (N00014-16-1-2007) 2020-08-19T14:15:08Z 2020-08-19T14:15:08Z 2019-05 2019-10-08T16:24:57Z Article http://purl.org/eprint/type/ConferencePaper 978-1-5386-6027-0 2577-087X https://hdl.handle.net/1721.1/126674 Ajay, Anurag et al. "Combining physical simulators and object-based networks for control." 2019 International Conference on Robotics and Automation (ICRA 2019), May 20-26, 2019, Montreal, Quebec: 3217-23 doi: 10.1109/ICRA.2019.8794358 ©2019 Author(s) en 10.1109/ICRA.2019.8794358 International Conference on Robotics and Automation (ICRA) Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf IEEE arXiv |
| spellingShingle | Ajay, Anurag. Bauza Villalonga, Maria Wu, Jiajun Fazeli, Nima Tenenbaum, Joshua B Rodriguez Garcia, Alberto Kaelbling, Leslie P Combining physical simulators and object-based networks for control |
| title | Combining physical simulators and object-based networks for control |
| title_full | Combining physical simulators and object-based networks for control |
| title_fullStr | Combining physical simulators and object-based networks for control |
| title_full_unstemmed | Combining physical simulators and object-based networks for control |
| title_short | Combining physical simulators and object-based networks for control |
| title_sort | combining physical simulators and object based networks for control |
| url | https://hdl.handle.net/1721.1/126674 |
| work_keys_str_mv | AT ajayanurag combiningphysicalsimulatorsandobjectbasednetworksforcontrol AT bauzavillalongamaria combiningphysicalsimulatorsandobjectbasednetworksforcontrol AT wujiajun combiningphysicalsimulatorsandobjectbasednetworksforcontrol AT fazelinima combiningphysicalsimulatorsandobjectbasednetworksforcontrol AT tenenbaumjoshuab combiningphysicalsimulatorsandobjectbasednetworksforcontrol AT rodriguezgarciaalberto combiningphysicalsimulatorsandobjectbasednetworksforcontrol AT kaelblinglesliep combiningphysicalsimulatorsandobjectbasednetworksforcontrol |