Researchers led by Gunhee Shin have built SE(3)-LIVOM, a system that estimates where a legged robot is and maps its surroundings, according to a paper on arXiv. It was the overall winner of the COMFORT Challenge at IROS 2026, which asks for the path of a legged robot in the GrandTour dataset, with scenes from collapsed buildings, construction sites and snowfields.
It works in real time with two lidars, a camera and an inertial measurement unit. Adaptive downsampling keeps the processing time near 35 ms per scan, and a back end corrects drift over time.
Using the same settings everywhere, its average absolute trajectory error across the six test scenes was 1.40 cm. Its code is available.



