FPGA design and implementation of a matrix multiplier based accelerator for 3D EKF SLAM

In hw/sw co-design FPGAs are being used in order to accelerate existing solutions so they meet real-time constraints. As they consume less power than a standard microprocessor and provide powerful parallel data processing capabilities, they remain a highly optimizable tool and object of research within an embedded system. In this paper we present an efficient architecture for matrix multiplication accelerator conceived as a systolic array co-processor to IBM's PPC440 processor on Virtex5 XC5VFX70T FPGA. Our design is afterwards synthesized and wired as a large-scale matrix multiplier required for an embedded version of a visual Simultaneous Localization and Mapping (SLAM) algorithm based on Extended Kaiman Filter (EKF). This algorithm is implemented entirely as a System On a programmable Chip (SoC) design on the FPGA; an EKF epoch is executed at least 7.3 times faster than the pure software implementation, maintaining and correcting 20 points in the map. This optimization permits an EKF block throughput to be increased from 6.07Hz to 44.39Hz, which exceeds our real-time constraint of 30Hz.