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

Abstract

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

A Universal LiDAR SLAM Accelerator System on Low-Cost FPGA

“…unless a good starting point is given, our previous accelerator only works in the frontend scan matching, where the initial guess is close to the globally optimal pose. In contrast, CSM core can handle backend loop detections with large initial errors as well as the frontend scan matching, which emphasizes the advantages of our proposal over [67] on both robustness and versatility. Our CSM core is able to produce results with the same quality as [66] using only one third the number of particles (50 and 16).…”

“…We also quote the results from various studies along with the execution environment, i.e., our previous work [67], GMapping with 50 particles [66], Google Cartographer [8], and Random Normal Matching [68], in Table 2. CSM core offers better accuracy than our previous work [67], where an FPGA-based accelerator for hill-climbing based scan matching is proposed and integrated to GMapping algorithm. Since hill-climbing method does not converge to a correct solution trans , and ε 2 rot are the accuracy metrics defined in Equations 10 and 11.…”

“…Comparison of the trajectory errors between PF-SLAM with the proposed FPGA accelerator (16 particles) and other recent works, i.e., GMapping with FPGA accelerator for hill-climbing based scan matching (32 particles) [67], GMapping (50 particles) [66], Google Cartographer [8], and Random Normal Matching [68].…”

An Extensive Analysis and Fine-Tuning of Gmapping’s Initialization Parameters

“…5 shows the visual description of a nine-square grid centered on the hit point, and neighboring unhit points are mainly utilized for the search box size when computing the score for the most likely laser beam strike in a kernel window. Parameters of Angle and linear displacement step increment size (lstep and astep) assumed the initial step size of the optimization step in the Hill-Climbing algorithm during the scan matching algorithm [28]. The Iterations parameter is the number of times the search step size changes during the optimization method in the scan matching algorithm.…”

An FPGA Overlay for Efficient Real-Time Localization in 1/10th Scale Autonomous Vehicles

“…K. Sugiura and H. Matsutani present an efficient hardware implementation for Simultaneous Localization and Mapping (SLAM) methods [17], sustaining a relatively small number of particles, despite a very efficient data compression strategy. The speed up the process of matching scanning on the FPGA in 2D LiDAR SLAM method called GMapping is based on the Rao-Blackwellized Particle Filter algorithm.…”