Development of GPU-accelerated localization system for autonomous mobile robot

Abstract

We present a localization system for autonomous mobile robot, that operates in conditions of well - known environment. In our work we use particle - based Monte - Carlo localization. This algorithm has many applications in mobile robotics, but it is computationally expensive. Due to high level of parallelism in this algorithm, we had an opportunity to accelerate its execution on graphical processing unit (GPU) using CUDA - parallel computing architecture from NVIDIA Corporation. We use Microsoft Kinect sensor and robot odometry to perform indoor localization. Developed system was tested on three-wheeled mobile robot. NVIDIA Jetson TK1 board performed all necessary computations.