EUVIMEDEUVIMED entdecken
Crossref

MIXL: Confidence-Aware Expert Fusion for Robust Multi-Sensor Localization

Keer Wang, Juliang Cao, Ruihang Yu, Zhiming Xiong, Meiping Wu

IEEE Robotics and Automation Letters · 2026-12 · 11 (12) · 14532-14539

DOI: 10.1109/lra.2026.3746083 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Robotics and Automation Letters, 2026-12. Verfasst von Keer Wang und 4 weiteren Autor:innen. Die Arbeit nennt 32 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Robust multi-sensor localization under sensor degradation requires reliable fusion decisions, not only fusion-weight estimation. Existing adaptive methods typically use soft weighting or hard expert selection from observation features, risking erroneous switching and trajectory drift when heterogeneous localization sources disagree. We propose MIXL, a confidence-aware fusion framework for heterogeneous localization sources. MIXL combines sensor-degradation cues with post-estimation source-health diagnostics, and uses a dual-head gate to separate source scoring from decision-confidence estimation. The estimated confidence controls sequential Commit, Hedge, and Defer execution. In Defer mode, MIXL retains the previous trusted routing state while all sources continue estimating and the system outputs the current pose. This policy suppresses unreliable instantaneous switching while preserving adaptation to heterogeneous degradation. Experiments on M2DGR and M3DGR under visual, LiDAR, GNSS, IMU, and multi-sensor degradation show competitive localization accuracy, calibrated decision confidence, and substantially improved temporal stability. On unseen M3DGR scenarios, MIXL achieves an average ATE of 0.16 m and only 1.15 expert switches per minute, substantially fewer than frame-wise hard selection.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Wang, K., Cao, J., Yu, R., Xiong, Z., & Wu, M. (2026). MIXL: Confidence-Aware Expert Fusion for Robust Multi-Sensor Localization. IEEE Robotics and Automation Letters, 11(12), 14532-14539. https://doi.org/10.1109/lra.2026.3746083

Vancouver

Wang K, Cao J, Yu R, Xiong Z, Wu M. MIXL: Confidence-Aware Expert Fusion for Robust Multi-Sensor Localization. IEEE Robotics and Automation Letters. 2026;11(12):14532-14539. doi:10.1109/lra.2026.3746083

BibTeX
@article{Wang2026,
  author = {Wang, Keer and Cao, Juliang and Yu, Ruihang and Xiong, Zhiming and Wu, Meiping},
  title = {MIXL: Confidence-Aware Expert Fusion for Robust Multi-Sensor Localization},
  journal = {IEEE Robotics and Automation Letters},
  year = {2026},
  volume = {11},
  number = {12},
  pages = {14532--14539},
  doi = {10.1109/lra.2026.3746083}
}
Kostenlos starten →