论文信息 - A FLOATING-POINTEXTENDED KALMAN FILTERIMPLEMENTATION FOR AUTONOMOUS MOBILEROBOTS

A FLOATING-POINTEXTENDED KALMAN FILTERIMPLEMENTATION FOR AUTONOMOUS MOBILEROBOTS

Localization andMapping aretwoofthemostimportant capabilities forautonomous mobile robots andhavebeenreceiving considerable attention fromthescientific computing community overthelast 10years. Oneofthemostefficient methods toaddress these problems isbased ontheuse oftheExtended KalmanFilter (EKF). TheEKFsimultaneously estimates amodeloftheenvironment (map)andthe position oftherobot based onodometric andexteroceptive sensor information. Asthis algorithm demands aconsiderable amountofcomputation, itisusually executed onhigh endPCscoupled totherobot. Inthis workwepresent an FPGA-based architecture fortheEKFalgorithm that iscapable ofprocessing two-dimensional mapscontaining upto 1.8k features atreal time(14Hz) andistwoorders ofmagnitudemorepowerefficient thanageneral purpose processor.

Eduardo Marques | Vanderlei Bonato

[1] Paolo Pirjanian,et al. Low-cost visual localization and mapping for consumer robotics , 2003, Ind. Robot.

[2] Randall Smith,et al. Estimating Uncertain Spatial Relationships in Robotics , 1987, Autonomous Robot Vehicles.

[3] Sebastian Thrun,et al. Probabilistic robotics , 2002, CACM.