CUDA Implementation of A Parallel Particle Filter for Mobile Robot Pose Estimation

Abstract

It is known as a challenging problem in the field of robot autonomous navigation to make navigation algorithms perform in real-time, including robot pose estimation, path planning, motion control and so on. One of sound solutions is to exploit hardware acceleration. Particle filter is a popular method for mobile robots pose estimation. The increase of the number of particles will improve the performance of the algorithm, which however brings detrimental effect on the real-time performance of the algorithm. GPU parallel computation is considered one of effective approaches to accelerate the computing speed of the particle filter algorithm. In this paper, a parallel particle filter algorithm for pose estimation of mobile robots is implemented using GPU acceleration and CUDA. Experiment results show that the approach can achieve an acceleration ratio as high as 17 in terms of the execution time of particle filter estimation algorithm.