Multi-thread impact on the performance of Monte Carlo based algorithms for self-localization of robots using RGBD sensors

Abstract

Abstract—Using information from RGBD sensors requires huge amount of processing. To use these sensors improves the robustness of algorithms for object perception, self-localization and, in general, all the capabilities to be performed by a robot to improve its autonomy. In most cases, these algorithms are not computationally feasible using single-thread implementations. This paper describes two multi thread strategies proposed for self localize a mobile robot in a known environment using information from a RGBD sensor. The experiments will show the benefits obtained when different numbers of threads are compared, using different approaches: a pool of threads and creation/destruction scheme. The work has been carried out on a Kobuki mobile robot in the environment of the RoCKiN competition, similar to RoboCup@home