certifiable-rwhe-calibration

Certifiably globally optimal generalized robot-world and hand-eye calibration. Our algorithm is the first extrinsic calibration method that can jointly solve for the poses of multiple sensors and targets, consider an unknown target scale, and provide a computational certificate of global optimality for its maximum likelihood objective function.

Usage

See experiments/rw_multi_eye_multi_hand.jl for an example of how to use our software on the processed real-world data in data/real-world/. Each .csv file in data/real-world/combine/ contains $A$ or $B$ measuerement matrices for the tag and camera indexed in its filename. For example, the $i$ th row of tag_0_cam_0_A.csv contains 7 floating point numbers representing a pose matrix $A_i \in \mathrm{SE}(3)$ matrix in an $A_i X_0 = Y_0 B_i$ measurement for tag 0 and camera 0: the first four are $w$, $x$, $y$, and $z$ of a unit quaternion representing a rotation, and the last three are a position in $\mathbb{R}^3$.

Citation

If you use this work in your research, please cite the following paper:

@article{wise2025certifably,
  title={A Certifably Correct Algorithm for Generalized Robot-World and Hand-Eye Calibration},
  author={Wise, Emmett and Kaveti, Pushyami and Chen, Qilong and Wang, Wenhao and Singh, Hanumant and Kelly, Jonathan and Rosen, David M and Giamou, Matthew},
  journal={arXiv preprint arXiv:2507.23045},
  year={2025}
}