Hochschule Karlsruhe Hochschule Karlsruhe - University of Applied Sciences
Hochschule Karlsruhe Hochschule Karlsruhe - University of Applied Sciences

IROS 2025: Joint paper with the Institute of Robotics and Intelligent Production Systems in Hangzhou, China

IROS 2025 - Yongzhou Zhang presented the paper "QBIT: Quality-Aware Cloud-Based Benchmarking for Robotic Insertion Tasks". The paper deals with a benchmarking framework for robot-based joining processes that evaluates not only the success rate but also quality metrics such as force application, force progression and execution time. To achieve realistic results, QBIT uses parameterized simulations, mesh decomposition, ROS2 containerization and a scalable Kubernetes intrastructure.

IRAS at the IROS 2025 in Hangzhou, China
- Presentation of the paper "QBIT: Quality-Aware Cloud-Based Benchmarking for Robotic Insertion Tasks"

At this year's IROS 2025 (IEEE/RSJ International Conference on Intelligent Robots and Systems) in Hangzhou, China, our PhD student Yongzhou Zhang presented the joint paper
"QBIT: Quality-Aware Cloud-Based Benchmarking for Robotic Insertion Tasks" together with Constantin Schempp from the Institute of Robotics and Intelligent Production Systems (IRP) .

The paper addresses a central question in applied robotics: How can robotic manipulation and insertion tasks be evaluated not only in terms of their success, but also in terms of execution quality? QBIT extends classic benchmarks, which usually only consider the success rate, to include further quality metrics such as precision, stability and execution efficiency.

Technically, QBIT is based on a modulable, cloud-based test setup that combines the following elements

  • Parameterized simulations to generate realistic test cases,
  • Mesh decomposition for detailed modeling of gripping and insertion geometries,
  • ROS2 containerization for reproducible, modular components,
  • and a scalable Kubernetes infrastructure for distributed and automatable benchmarking.

This integrated approach makes it possible to generate repeatable, realistic and comparable test scenarios, both for research purposes and for evaluating industrial solutions. The cloud architecture also enables easy reproducibility and scaling of the experiments.

We congratulate Yongzhou Zhang and Constantin Schempp on their successful presentation.

Paper: https://arxiv.org/abs/2503.07479