Accuracy improvement of cooperative localization using UAV and UGV

In a simultaneous localization and mapping (SLAM) system, in general, distribution estimation based on a particle filter is used for localization of Unmanned Ground Vehicle (UGV) with a laser rangefinder, and point estimation based on optimization is used for Unmanned Aerial Vehicle (UAV) with a monocular camera. In order for such robots to perform cooperative localization to improve accuracy, it is necessary to convert the localization results of point estimation and distribution estimation into a format that can be introduced to each other. In this paper, we propose an integration method of estimation results for cooperative localization by UAV and UGV. The UAV point estimation result is converted to a normal distribution by giving a fixed covariance matrix in order to introduce it as the observation likelihood of the UGV. On the other hand, the particle pose with maximum weight obtained from the UGV distribution estimation result is added to the current pose constraint in UAV optimization. As a result, this allows stable localization of both by correcting with the estimation result of the other even in an environment where either one is disadvantageous for localization.

Automated summary

This paper proposes an integration method of estimation results for cooperative localization by UAV and UGV. The point estimation result of the UAV is converted to a normal distribution and introduced as the observation likelihood of the UGV, while the particle pose with maximum weight from the UGV's distribution estimation result is added to the current pose constraint in UAV optimization. The results show that stable localization of both robots can be achieved by correcting with the estimation result of the other even in an environment where either one is disadvantageous for localization.