Simultaneous Localization and Mapping (SLAM) is a key challenge in mobile
robotics. Single-robot SLAM that enables one robot to localize within a self-
build map has reached a matured level and commercial products based on this
technology have settled on the market. So Multi-Robot SLAM (MRSLAM)
that enables a team of robots to map an area is getting more and more interest
and showing some promising results.
This degree project, conducted at Inkonova AB, proposes a LiDAR-based
MRSLAM where multiple robots can individually build their own local maps
before sharing and merging them when meeting each other. The method is
decentralized to enable the team to map complex environment without the
need of infrastructure or assuming any continuous connection between them,
scenarios that are commonly found in underground areas. Each robot builds
its local map using LiDAR Odometry And Mapping (LOAM) and aligns maps
from other robots using a two-steps method: a feature-based algorithm aligns
roughly the maps that is then refined using Iterative Closest Point (ICP).
The proposed solution is evaluated through four experiments focusing firstly
on assessing some specific modules before testing the full method. The results
show that the MRSLAM system successfully maps an area by allowing robots
to detect each other, and to match their maps before merging them.
Keywords: Multi-Robot SLAM, LiDAR, 3D collaborative mapping,
drone