An Optimization of Multi Sensor SLAM Algorithm for SIMD Architecture

Abstract

This thesis aims to optimize movement and localization of robot using algorithms designed newlinefor GP-GPU architecture; for which various Simultaneous Localization and Mapping (SLAM) newlinealgorithms had studied like FastSLAM, Extended Kalman Filter based SLAM, Range Only newlineSLAM, etc. To test parallelism in these algorithms various components were deployed like newlinethe 3D Accelerometer Sensor of eZ430 Chronos Texas Instrument Watch. It has been found newlinethat Extended Kalman Filter (EKF)based SLAM gives optimized and scalable results. newlineFor the design and implementation of EKF based SLAM algorithm, we leverage AMD newlineRadeon, serial adapter, and OpenCL API on Graphics Core Next based system. To achieve newlinethis, various memory models had studied like Uniform Memory Access (UMA), Non- newlineUniform Memory Access (NUMA), Cache Coherence Uniform Memory Access (CCUMA), newlineand it has been observed that GP-GPU models can be applied for coherence and localization newlineof memory for Extended Kalman Filter, which uses Matrix Multiplication in near branch newlinedata. As this model is based on Spatial Locality of Reference the CCUMA memory is the newlinebest candidate model. Also, the basic model was determined along with the parameter needed newlineduring navigation of the robot. Here, testing and optimization of the presented EKF based newlineSLAM algorithm take place on Arduino simulators on .Net platform with non-visual sensors. newlineThese sensors are used to predict the next step taken by the robot in an environment which newlinecan be improved further using curve SLAM algorithms. Performance analysis of the robot newlinewas done using Sectorial Error Probability(SEP), it was found that SEP was not optimal so newlinewe tried parametric curve algorithm and found suitability of Curve SLAM techniques (Bezier newlineCurve, BSpline) for smoothing the path of the robot. The designed robot is equipped with newlinemulti-sensors viz: IMU, SONAR; the IMU is used to determine environment and localization newlinewhich is subtended to parametric Bezier and BSpline for better path estimation.