Efficient GPU-based construction of occupancy grids using several laser range-finders

Abstract

Building occupancy grids (OGs) in order to model the surrounding environment of a vehicle implies to fusion occupancy information provided by the different embedded sensors in the same grid. The principal difficulty comes from the fact that each can have a different resolution, but also that the resolution of some sensors varies with the location in the field of view. In this article we present a new efficient approach to this issue based upon a graphical processor unit (GPU). In that perspective, we explain why the problem of switching coordinate systems is an instance of the texture mapping problem in computer graphics. We also present an exact algorithm in order to evaluate the accuracy of such a device, which is not precisely known due to the several approximations made by the hardware. To validate our method, the results with GPU are also compared to results obtained through the exact approach and the GPU precision is shown to be good enough for robotic applications. Therefore we describe a whole and general calculus architecture to build occupancy grids for any kind of range-finder with a graphical processor unit (GPU). And we present computational time results that can allow to compute occupancy grids for 50 sensors at frame rate even for a very fine grid.

Fast Processing of Grid Maps using Graphical Multiprocessors

“…It is a bit surprising that after the performance shown in Yguel et al (2008), the use of GPUs for processing in mobile robotics has not been widely adopted by the community. We believe that mobile robotics algorithms necessarily have to be adapted for heterogeneous (CPU and GPU) multi-core processors to benefit from modern hardware computational power and to be able to increase their performance up to the next level.…”

“…There exists a very closely related work from Yguel et al (2008), in which GPU based processing of grid maps is also presented. Their work claims that grid mapping algorithms based on ray tracing, commonly using the line tracing algorithm of Bresenham (1965) are inaccurate as they present the de Moire effect, defined in the computational graphics domain as artificial discontinuities between rays far from the origin.…”

“…Furthermore, the full grid mapping algorithm is addressed to the GPU, as opposed to Yguel et al (2008) in which only certain parts of the algorithm are processed in the GPU, so we are able to present absolute timings comparison between the same algorithm in a CPU implementation and a GPU one. The main contribution of the paper is the proof that exactly the same algorithm that is being used by a large part of the community can be targeted to a graphical card with large computational savings and fully releasing the CPU.…”

Dense Mapping for Range Sensors: Efficient Algorithms and Sparse Representations

“…Defining these functions is an important part of many works ( [8], [19]). The results in [20] prove that for certain choices of parameters 3 these functions are piecewise constants:…”

GPSlam: Marrying Sparse Geometric and Dense Probabilistic Visual Mapping

“…Other approaches focus on building highly efficient grid mapping algorithms. To allow for real-time updates, GPU-implementations [16] or multiresolution approaches [2] are proposed. Others focus on reducing memory consumption in 3D mapping, with multi-volume grids [6] or octree representations [30].…”