This work proposes a novel physics-based Cartpole simulation environment as a new benchmark to address the sim-to-real transfer. Our simulation environment extends the original Gymnasium Cartpole environment with additional physics and data-driven models for friction, air resistance, and the nonlinear behavior of the applied force on the cart inspired by a real-world experimental setup. We implement the Gymnasium environment interface, allowing us to use our implementation as a drop-in replacement with configurable simulation fidelity. We show that our physics-based Cartpole simulation with Reinforcement Learning minimizes the reality gap to our real-world Cartpole setup without increasing computational efforts considerably. Moreover, our simulation environment is an efficient surrogate model for a real Cartpole, and thus provides a rare example of closing the reality gap.
Description
High-Fidelity Simulation of a Cartpole for Sim-to-Real Deep Reinforcement Learning | IEEE Conference Publication | IEEE Xplore
%0 Conference Paper
%1 10602859
%A Bantel, Linus
%A Domanski, Peter
%A Pflüger, Dirk
%B 2024 4th Interdisciplinary Conference on Electrics and Computer (INTCEC)
%D 2024
%K
%P 1-6
%R 10.1109/INTCEC61833.2024.10602859
%T High-Fidelity Simulation of a Cartpole for Sim-to-Real Deep Reinforcement Learning
%X This work proposes a novel physics-based Cartpole simulation environment as a new benchmark to address the sim-to-real transfer. Our simulation environment extends the original Gymnasium Cartpole environment with additional physics and data-driven models for friction, air resistance, and the nonlinear behavior of the applied force on the cart inspired by a real-world experimental setup. We implement the Gymnasium environment interface, allowing us to use our implementation as a drop-in replacement with configurable simulation fidelity. We show that our physics-based Cartpole simulation with Reinforcement Learning minimizes the reality gap to our real-world Cartpole setup without increasing computational efforts considerably. Moreover, our simulation environment is an efficient surrogate model for a real Cartpole, and thus provides a rare example of closing the reality gap.
@inproceedings{10602859,
abstract = {This work proposes a novel physics-based Cartpole simulation environment as a new benchmark to address the sim-to-real transfer. Our simulation environment extends the original Gymnasium Cartpole environment with additional physics and data-driven models for friction, air resistance, and the nonlinear behavior of the applied force on the cart inspired by a real-world experimental setup. We implement the Gymnasium environment interface, allowing us to use our implementation as a drop-in replacement with configurable simulation fidelity. We show that our physics-based Cartpole simulation with Reinforcement Learning minimizes the reality gap to our real-world Cartpole setup without increasing computational efforts considerably. Moreover, our simulation environment is an efficient surrogate model for a real Cartpole, and thus provides a rare example of closing the reality gap.},
added-at = {2024-09-18T12:20:58.000+0200},
author = {Bantel, Linus and Domanski, Peter and Pflüger, Dirk},
biburl = {https://puma.ub.uni-stuttgart.de/bibtex/2555d7e2e44897b72e80d7fd6c3455c70/lbantel},
booktitle = {2024 4th Interdisciplinary Conference on Electrics and Computer (INTCEC)},
description = {High-Fidelity Simulation of a Cartpole for Sim-to-Real Deep Reinforcement Learning | IEEE Conference Publication | IEEE Xplore},
doi = {10.1109/INTCEC61833.2024.10602859},
interhash = {3cdad7c5c9334d7de3aca31ab249a26e},
intrahash = {555d7e2e44897b72e80d7fd6c3455c70},
keywords = {},
month = {June},
pages = {1-6},
timestamp = {2024-09-18T12:20:58.000+0200},
title = {High-Fidelity Simulation of a Cartpole for Sim-to-Real Deep Reinforcement Learning},
year = 2024
}
