Analysis of Randomization Effects on Sim2Real Transfer in Reinforcement Learning for Robotic Manipulation Tasks
Abstract
Current state-of-the-art approaches for transferring deep-learning models trained in simulation either rely on highly realistic simulations or employ randomization techniques to bridge the reality gap. However, such strategies do not scale well for complex robotic tasks. Highly-realistic simulations are computationally expensive and challenging to implement, while randomization techniques become sample-inefficient as the complexity of the task increases. This paper proposes a procedure for training on incremental simulations in a continual learning setup. We develop a simulation platform for the experimental analysis that can serve as a training environment and as a benchmark for continual and reinforcement learning sim2real approaches. The results show that training time for complex tasks can be reduced. Thus, we argue that Sequentially-Randomized Simulations improve the sim2real transfer.
Files
Cite and export
Choose a format, then copy or download.
@inproceedings{Josifovski2022Analysis,
author = {Josifovski, Josip and Malmir, Mohammadhossein and Klarmann, Noah and Zagar, Bare Luka
and Navarro-Guerrero, Nicolás and Knoll, Alois},
title = {Analysis of Randomization Effects on {Sim2Real} Transfer in Reinforcement Learning
for Robotic Manipulation Tasks},
booktitle = {{IEEE}/{RSJ} International Conference on Intelligent Robots and Systems ({IROS})},
pages = {10193--10200},
month = oct,
year = {2022},
publisher = {IEEE},
address = {Kyoto},
isbn = {978-1-66547-927-1},
doi = {10.1109/IROS47612.2022.9981951},
url = {https://robotik.dfki-bremen.de/en/research/publications/12962}
}
TY - CPAPER AU - Josifovski, Josip AU - Malmir, Mohammadhossein AU - Klarmann, Noah AU - Zagar, Bare Luka AU - Navarro-Guerrero, Nicolás AU - Knoll, Alois TI - Analysis of Randomization Effects on Sim2Real Transfer in Reinforcement Learning for Robotic Manipulation Tasks T2 - IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) SP - 10193 EP - 10200 PY - 2022 DA - 2022/10// PB - IEEE CY - Kyoto SN - 978-1-66547-927-1 DO - 10.1109/IROS47612.2022.9981951 AB - Current state-of-the-art approaches for transferring deep-learning models trained in simulation either rely on highly realistic simulations or employ randomization techniques to bridge the reality gap. However, such strategies do not scale well for complex robotic tasks. Highly-realistic simulations are computationally expensive and challenging to implement, while randomization techniques become sample-inefficient as the complexity of the task increases. This paper proposes a procedure for training on incremental simulations in a continual learning setup. We develop a simulation platform for the experimental analysis that can serve as a training environment and as a benchmark for continual and reinforcement learning sim2real approaches. The results show that training time for complex tasks can be reduced. Thus, we argue that Sequentially-Randomized Simulations improve the sim2real transfer. UR - https://robotik.dfki-bremen.de/en/research/publications/12962 LA - eng ER -
Josifovski, J., Malmir, M., Klarmann, N., Zagar, B. L., Navarro-Guerrero, N., & Knoll, A. (2022). Analysis of Randomization Effects on Sim2Real Transfer in Reinforcement Learning for Robotic Manipulation Tasks. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (pp. 10193–10200). IEEE. https://doi.org/10.1109/IROS47612.2022.9981951
Back to the list