An Analysis of Parallel Approaches for a Mobile Robotic Self-localization Algorithm

摘要

Self-localization is a fundamental problem in mobile robotics. It consists of estimating the position of a robot given a map of the environment and information obtained by sensors. Among the algorithms used to address this issue, the Monte Carlo technique has obtained a considerable attention by the scientific community due to its simplicity and efficiency. Monte Carlo localization is a sample-based technique that estimates robot´s pose using a probability density function represented by samples (particles). The complexity of this algorithm scales proportionally to the number of particles used. The larger the environment, the more particles are required for robot localization. This fact limits the use of this algorithm in large size environments. In order to improve the efficiency of the Monte Carlo technique and allow it to be used in large environments we propose a parallel implementation of it. Our implementation is based on OpenMP and MPI message passing interface. Experimental results are used to show the efficiency of our approach.