Detecting other robots from range data for robust self-localization of autonomous mobile robots

Abstract

We have developed autonomous mobile robots that delivery products in factories or warehouses while recognizing their own positions using bearing-range sensors. For robust self-localization of such robots in environments that include multiple robots, we propose a technique for correctly detecting other robots' surface points from range data. In these environments, conventional robots cannot correctly distinguish the measured points of other robots' surfaces from those of walls until they have achieved self-localization even if the robots are informed of the rough positions of the other robots. This may lead to non-negligible errors in the robot's self-localization. To solve this problem, our technique accurately estimates other robots' positions by locally maximizing their likelihoods calculated on the areas around the rough positions of other robots using range data and the robots' shapes that are preliminarily known. Furthermore, every measured point is classified as either another robot's point or a general point according to the deviation between the actual measured distance and the virtual distance calculated using the accurately estimated positions of other robots. We conducted a simulation and verified that with our technique, other robots' points were detected from range data with more than a 99% success rate on the condition that the deviation tolerance was 5 cm, which is five times that of the standard deviation of measured distance errors. When our technique was implemented in robots, they were able to robustly recognize their own positions using the range data from which the measured points of other robots' surfaces had been eliminated.