3D/6DOF Particle Filtering Location Algorithm Based on GPU Parallel Acceleration

Autonomous navigation ability is considered to be one of the important functions of intelligent mobile robot system. The localization method based on particle filter is widely used in the localization field of mobile robot. Under the background of traditional single thread computing method, the positioning accuracy and real-time performance of particle filter location algorithm are often difficult to be guaranteed simultaneously. To solve this problem, we propose to use graphics processing units (GPU) to carry out large-scale parallel computation on particles, so as to achieve the acceleration effect. Specifically, we analyse the parallelism of the prediction, observation and resampling aspects of the mcl_3dl algorithm, and do parallel acceleration of the particle weight update in the observation part. The acceleration effect of the proposed algorithm was tested in pose tracking experiments and global localization experiments. The results show that the particle weight updating speed of the algorithm in indoor scene and outdoor scene is improved by 16.7% and 37.8%, respectively. In the indoor corridor scene, the convergence speed of the global localization is increased by 29.1%.