Robot Pouring: Modeling and Sim-to-Real Evaluation Using Causal Discovery

Maldonado, Jaime; Huerkamp, Malte; Krumme, Jonas; Zetzsche, Christoph; Beetz, Michael

Robot Pouring: Modeling and Sim-to-Real Evaluation Using Causal Discovery

by Jaime Maldonado, Malte Huerkamp, Jonas Krumme, Christoph Zetzsche, Michael Beetz

Abstract:

Robot simulations constitute a practical option for generating large training datasets. However, simulations based on simplified models of physical phenomena might deviate from the behavior of the real system. In this paper, we use causal analysis methods to identify and evaluate the mismatch between simulation and reality of a robot pouring task. The cause-effect structure of the simulation was identified using a causal discovery algorithm. The discovered structure was evaluated against data acquired in a real robot setup. As a result, a factor of simulation-to-real mismatch was identified. The practical use of causal discovery for simulation-based modeling is demonstrated.

Reference:

Robot Pouring: Modeling and Sim-to-Real Evaluation Using Causal Discovery (Jaime Maldonado, Malte Huerkamp, Jonas Krumme, Christoph Zetzsche, Michael Beetz), In European Robotics Forum 2025 (Marco Huber, Alexander Verl, Werner Kraus, eds.), Springer Nature Switzerland, 2025.

Bibtex Entry:

@InProceedings{maldonado2025pouringsimtoreal,
author={Maldonado, Jaime and Huerkamp, Malte and Krumme, Jonas and Zetzsche, Christoph and Beetz, Michael},
editor={Huber, Marco and Verl, Alexander and Kraus, Werner},
title={Robot Pouring: Modeling and Sim-to-Real Evaluation Using Causal Discovery},
booktitle={European Robotics Forum 2025},
year={2025},
publisher={Springer Nature Switzerland},
address={Cham},
pages={314--319},
abstract={Robot simulations constitute a practical option for generating large training datasets. However, simulations based on simplified models of physical phenomena might deviate from the behavior of the real system. In this paper, we use causal analysis methods to identify and evaluate the mismatch between simulation and reality of a robot pouring task. The cause-effect structure of the simulation was identified using a causal discovery algorithm. The discovered structure was evaluated against data acquired in a real robot setup. As a result, a factor of simulation-to-real mismatch was identified. The practical use of causal discovery for simulation-based modeling is demonstrated.},
isbn={978-3-031-89471-8},
keywords = {EASE-H1}
}