General Purpose Computing on Graphics Processing Units for Robotic Applications

This paper deals with research related with the improvements of state of the art algorithms used in robotic applications based on parallel computation. The main goal is to decrease the computational complexity of 3D cloud of points processing in such applications as: data filtering, normal vector estimation, data registration and calculation of point feature histogram. The presented results efficiently improve the existing implementations with minimal lost of accuracy. The main contribution is a regular grid decomposition originally implemented for nearest neighborhood search. This data structure is used almost for all presented methods, it provides an efficient method for decreasing the time of computation. The results are compared with well-known robotic frameworks such as PCL and 3DTK.