A Differentiable Physics Engine for Deep Learning in Robotics
An important field in robotics is the optimization of controllers. Currently, robots are often treated as a black box in this optimization process, which is the reason why derivative-free optimization methods such as evolutionary algorithms or reinforcement learning are omnipresent. When gradient-ba...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2019-03-01
|
| Series: | Frontiers in Neurorobotics |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/article/10.3389/fnbot.2019.00006/full |
| _version_ | 1819225727340380160 |
|---|---|
| author | Jonas Degrave Michiel Hermans Joni Dambre Francis wyffels |
| author_facet | Jonas Degrave Michiel Hermans Joni Dambre Francis wyffels |
| author_sort | Jonas Degrave |
| collection | DOAJ |
| description | An important field in robotics is the optimization of controllers. Currently, robots are often treated as a black box in this optimization process, which is the reason why derivative-free optimization methods such as evolutionary algorithms or reinforcement learning are omnipresent. When gradient-based methods are used, models are kept small or rely on finite difference approximations for the Jacobian. This method quickly grows expensive with increasing numbers of parameters, such as found in deep learning. We propose the implementation of a modern physics engine, which can differentiate control parameters. This engine is implemented for both CPU and GPU. Firstly, this paper shows how such an engine speeds up the optimization process, even for small problems. Furthermore, it explains why this is an alternative approach to deep Q-learning, for using deep learning in robotics. Finally, we argue that this is a big step for deep learning in robotics, as it opens up new possibilities to optimize robots, both in hardware and software. |
| first_indexed | 2024-12-23T10:14:11Z |
| format | Article |
| id | doaj.art-dd822b600f614bcda2d1dfff239a2499 |
| institution | Directory Open Access Journal |
| issn | 1662-5218 |
| language | English |
| last_indexed | 2024-12-23T10:14:11Z |
| publishDate | 2019-03-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Neurorobotics |
| spelling | doaj.art-dd822b600f614bcda2d1dfff239a24992022-12-21T17:50:52ZengFrontiers Media S.A.Frontiers in Neurorobotics1662-52182019-03-011310.3389/fnbot.2019.00006406386A Differentiable Physics Engine for Deep Learning in RoboticsJonas Degrave0Michiel Hermans1Joni Dambre2Francis wyffels3IDLab-AIRO, Department of Electronics and Information Systems, Ghent University - imec, Ghent, BelgiumIndependent Researcher, Ghent, BelgiumIDLab-AIRO, Department of Electronics and Information Systems, Ghent University - imec, Ghent, BelgiumIDLab-AIRO, Department of Electronics and Information Systems, Ghent University - imec, Ghent, BelgiumAn important field in robotics is the optimization of controllers. Currently, robots are often treated as a black box in this optimization process, which is the reason why derivative-free optimization methods such as evolutionary algorithms or reinforcement learning are omnipresent. When gradient-based methods are used, models are kept small or rely on finite difference approximations for the Jacobian. This method quickly grows expensive with increasing numbers of parameters, such as found in deep learning. We propose the implementation of a modern physics engine, which can differentiate control parameters. This engine is implemented for both CPU and GPU. Firstly, this paper shows how such an engine speeds up the optimization process, even for small problems. Furthermore, it explains why this is an alternative approach to deep Q-learning, for using deep learning in robotics. Finally, we argue that this is a big step for deep learning in robotics, as it opens up new possibilities to optimize robots, both in hardware and software.https://www.frontiersin.org/article/10.3389/fnbot.2019.00006/fulldifferentiable physics enginedeep learninggradient descentneural network controllerrobotics |
| spellingShingle | Jonas Degrave Michiel Hermans Joni Dambre Francis wyffels A Differentiable Physics Engine for Deep Learning in Robotics Frontiers in Neurorobotics differentiable physics engine deep learning gradient descent neural network controller robotics |
| title | A Differentiable Physics Engine for Deep Learning in Robotics |
| title_full | A Differentiable Physics Engine for Deep Learning in Robotics |
| title_fullStr | A Differentiable Physics Engine for Deep Learning in Robotics |
| title_full_unstemmed | A Differentiable Physics Engine for Deep Learning in Robotics |
| title_short | A Differentiable Physics Engine for Deep Learning in Robotics |
| title_sort | differentiable physics engine for deep learning in robotics |
| topic | differentiable physics engine deep learning gradient descent neural network controller robotics |
| url | https://www.frontiersin.org/article/10.3389/fnbot.2019.00006/full |
| work_keys_str_mv | AT jonasdegrave adifferentiablephysicsenginefordeeplearninginrobotics AT michielhermans adifferentiablephysicsenginefordeeplearninginrobotics AT jonidambre adifferentiablephysicsenginefordeeplearninginrobotics AT franciswyffels adifferentiablephysicsenginefordeeplearninginrobotics AT jonasdegrave differentiablephysicsenginefordeeplearninginrobotics AT michielhermans differentiablephysicsenginefordeeplearninginrobotics AT jonidambre differentiablephysicsenginefordeeplearninginrobotics AT franciswyffels differentiablephysicsenginefordeeplearninginrobotics |