On the robot path planning using cloud computing for large grid maps

Abstract

Global path planning consists in finding the optimal path for a mobile robot with the lowest cost in the minimum amount of time, without colliding with the obstacles scattered in the workspace. In this paper, we investigate the benefits of offloading path planning algorithms to be executed in the cloud rather than in the robot. The contribution consists in developing a vertex-centric implementation of RA* [1], a version of A* that we developed for grid maps and that was proven to be much faster than A*, using the distributed graph processing framework Giraph that rely on Hadoop. We also developed a centralized cloud-based C++ implementation of the algorithm for benchmarking and comparison purposes. Experimental results on a real cloud shows that the distributed graph processing Giraph fails to provide faster execution as compared to centralized C++ implementation for different map sizes and configuration due to non-real time properties of Hadoop.