A High Speed and Low Complexity Architecture Design Methodology for Square Root Unscented Kalman Filter based SLAM

Square Root Unscented Kalman Filter (SRUKF) provides a practical solution for highly nonlinear and critical applications such as Simultaneous Localization and Mapping (SLAM). It improves the stability of the system, at the same time enhancing the numerical accuracy for hardware platform. It is less computationally demanding than the Unscented Kalman Filter (UKF), however, the hardware development process remains a complex task with high resource utilization. This paper presents a Householder CORDIC based low complexity architecture design for SRUKF targeting the highly nonlinear SLAM application. A hardware-software co-design methodology is proposed and implemented on Zynq-7000 XC7Z020 FPGA. Synthesis results show that the architecture saves 91% of DSP cores and is 20% faster than the state-of-the-art UKF architecture. The proposed SRUKF is also highly stable and achieve 78% and 10% greater accuracy than EKF SLAM and UKF SLAM respectively.