Super Odometry 2.0: reliable robot localisation in smoke, snow and dust

In a paper posted to arXiv on 26 August 2026, Shibo Zhao and six co-authors address a hard problem for robots: when cameras or lidar are degraded by smoke, snow, dust or low light, localisation reliability drops. The record references a 2025 Science Robotics publication.

Their contribution is hierarchical sensor fusion with adaptive feature selection, state-direction estimation, engine selection and learned inertial odometry. The authors report validation over 200 km and 800 operational hours.

The value lies in graceful degradation and sensor fallback. The results are author-reported; operating conditions, baselines and reproduction need to be examined in the full paper, and the timing gap between the journal reference and the arXiv posting is worth noting.