A new article, “Long-term LiDAR Localization in Dynamic Environments: Comparison with OSS Using a Public Benchmark” has been published on Kudan Tech Blog.
At construction sites, stockpiled materials, scaffolding, traffic cones, and temporary walkways move on a daily basis, and as construction progresses, new walls appear week by week. As a result, maps created in advance for robots drift out of line with the actual state of the site day by day. In this article, Kudan compared how stably its LiDAR localization module, currently under development, operates in such changing environments with leading open-source implementations, using public datasets. The proportion of time within an error of 5 cm was approximately 1–47% for the open-source implementations, compared with approximately 87% for Kudan’s module (supporting up to 99% in practical use). Kudan also confirmed that, even under conditions where the map and the site differ significantly, the robot can keep moving by updating the map automatically.
The challenge in using robots at construction sites is the effort of rebuilding maps to keep up with a constantly changing site so that the robots can keep operating. Kudan’s localization technology is being developed in three directions to address this challenge.
First, building maps from existing digital twins such as BIM and point clouds, eliminating the need for dedicated map-building runs by robots.
Second, continuing to run without losing position even when the placement of materials and temporary structures changes.
Third, automatically reflecting changes detected during operation in the map.
Through these, Kudan aims to build a foundation that allows robots to operate without stopping, even at sites where the environment keeps changing as construction progresses, such as for material transport, patrol inspection, and construction progress checks. Going forward, Kudan will continue to contribute to the expansion of robot use through practical deployment at sites with significant change, starting with construction.
Article: “Long-term LiDAR Localization in Dynamic Environments: Comparison with OSS Using a Public Benchmark”
https://zenn.dev/kudan_jp/articles/ceeac0e3d15344?locale=en
Kudan Tech Blog (Japanese)
https://zenn.dev/p/kudan_jp
