An FPGA Acceleration and Optimization Techniques for 2D LiDAR SLAM Algorithm

An efficient hardware implementation for Simultaneous Localization and Mapping (SLAM) methods is of necessity mobile autonomous robots with limited computational resources. In this paper, we propose a resource-efficient FPGA accelerating scan matching computations, which typically cause major bottleneck in 2D LiDAR SLAM methods. Scan process correcting robot pose by aligning the latest measurements an occupancy grid map, encodes information about surrounding environment. We exploit inherent parallelism Rao-Blackwellized Particle Filter (RBPF) based algorithm to perform computations multiple particles parallel. proposed design, several techniques are employed reduce resource utilization achieve maximum throughput. Experimental results using benchmark datasets show that accelerated 5.31-8.75× overall throughput improved 3.72-5.10× without seriously degrading quality final outputs. Furthermore, our IP core requires only 44% total resources available TUL Pynq-Z2 board, thus facilitating realization applications on indoor robots.