This site complements the paper Bayesian In-Situ Joint-Torque Sensor Calibration for Robotic Hands via Self-Consistency under Contact Constraints by Leon Sievers and Berthold Bäuml presented at the 2026 IEEE-RAS International Conference on Humanoid Robots in San Jose, CA, USA.

Abstract

---

Dexterous manipulation with multi-fingered robotic hands depends on the precise control of contact forces. For torque-controlled hands, achieving the necessary reliability and—especially for learning in simulation—successful sim-to-real transfer relies heavily on accurately calibrated torque sensing. Conventional calibration methodologies, however, are notoriously tedious, often demanding specialized external measurement hardware and partial system disassembly. This paper introduces a novel, autonomous approach that facilitates the in-situ calibration of a complete multi-fingered hand in a matter of minutes, simply through pairwise fingertip contacts in multiple configurations. The proposed framework exploits the fact that a low-dimensional contact wrench induces a self-consistency constraint on the joint torques of the involved fingers, completely bypassing the need to know the wrench’s actual value. Based on a Bayesian formulation, we derive a robust calibration method and provide a comprehensive observability analysis. In experiments with the DLR Hand II, our automated procedure achieves a precision comparable to the manual gold standard—which requires disassembling the hand—resulting in a force error of only 0.39N at the fingertip.

---

Cite this paper as:

@inproceedings{Sievers2026,
    author = {Leon Sievers and Berthold B{\"a}uml},
    title = {Bayesian In-Situ Joint-Torque Sensor Calibration for Robotic Hands via Self-Consistency under Contact Constraints},
    booktitle = {IEEE-RAS International Conference on Humanoid Robots (Humanoids)},
    year = {2026}}

arxiv