Manual development of control systems’ software is time-consuming and error-prone. Thus, high costs are already incurred in this phase of mechatronic system development. Virtual prototypes have so far only been used for testing purposes, such as virtual commissioning, but not for the automated creation of the control. A good test environment can also be extended to a learning environment with appropriate trial and error based algorithms. This work shows an approach to extend an industrial software tool for virtual commissioning as a standardized OpenAI gym environment. Thereby, established reinforcement learning algorithms can be used more easily and a step towards an industrial application of self-learning control systems can be made. The goal of this work is to provide industry and research with a platform for easy entry into the field of reinforcement learning.
%0 Conference Paper
%1 10123668
%A Jaensch, Florian
%A Klingel, Lars
%A Verl, Alexander
%B 2022 5th International Conference on Artificial Intelligence for Industries (AI4I)
%D 2022
%K commissioning isw learning myown virtual
%P 64-67
%R 10.1109/AI4I54798.2022.00023
%T Virtual Commissioning Simulation as OpenAI Gym - A Reinforcement Learning Environment for Control Systems
%X Manual development of control systems’ software is time-consuming and error-prone. Thus, high costs are already incurred in this phase of mechatronic system development. Virtual prototypes have so far only been used for testing purposes, such as virtual commissioning, but not for the automated creation of the control. A good test environment can also be extended to a learning environment with appropriate trial and error based algorithms. This work shows an approach to extend an industrial software tool for virtual commissioning as a standardized OpenAI gym environment. Thereby, established reinforcement learning algorithms can be used more easily and a step towards an industrial application of self-learning control systems can be made. The goal of this work is to provide industry and research with a platform for easy entry into the field of reinforcement learning.
@inproceedings{10123668,
abstract = {Manual development of control systems’ software is time-consuming and error-prone. Thus, high costs are already incurred in this phase of mechatronic system development. Virtual prototypes have so far only been used for testing purposes, such as virtual commissioning, but not for the automated creation of the control. A good test environment can also be extended to a learning environment with appropriate trial and error based algorithms. This work shows an approach to extend an industrial software tool for virtual commissioning as a standardized OpenAI gym environment. Thereby, established reinforcement learning algorithms can be used more easily and a step towards an industrial application of self-learning control systems can be made. The goal of this work is to provide industry and research with a platform for easy entry into the field of reinforcement learning.},
added-at = {2023-06-28T10:18:24.000+0200},
author = {Jaensch, Florian and Klingel, Lars and Verl, Alexander},
biburl = {https://puma.ub.uni-stuttgart.de/bibtex/2623843e14e08724a1635c6202bd3573f/isw-bibliothek},
booktitle = {2022 5th International Conference on Artificial Intelligence for Industries (AI4I)},
doi = {10.1109/AI4I54798.2022.00023},
interhash = {aa3a16798397cb577c368777d54cafed},
intrahash = {623843e14e08724a1635c6202bd3573f},
issn = {2770-4718},
keywords = {commissioning isw learning myown virtual},
month = {Sep.},
pages = {64-67},
timestamp = {2023-06-28T10:18:24.000+0200},
title = {Virtual Commissioning Simulation as OpenAI Gym - A Reinforcement Learning Environment for Control Systems},
year = 2022
}
