Sensor Calibration Using theNeural Extended KalmanFilter inaControl Loop

Abstract

Sensor errors canadversely affect thebehavior ofa atthenecessary update rates. Insuchcases, less accurate control system. Whenmultiple sensors areused, abroken sensorsensors mightbeusedtoprovide reports atthesampling time canhaveitseffects minimized byartificially inflating itserrorbetween themeasurement reports ofthemoreaccurate covariance. Inthis paper, adifferent approach tocompensatingfor system. Finally, additional sensors areaddedinthat they sensor errors inamultiple-sensor control system isintroduced The provide measurements thatcontain directly observable state technique, referred toasaneural extended Kalmanfilter (NEKF), is information. Inmobile robotic systems, forexample, sensors developed forclosed-loop control systems. TheNEKFlearns on- include.posIn sors todetermis,or e plesuch line from thesameresidual information usedinthestate estimator. include position sensors todetermine absolute position, such Theimprovement inthesensor report ismadebytheneural networkasa GPS,orrelative position suchasradars, sonars and being added tothemeasurement model. Inthis work, theNEKFis imagery equipment, aswellasspeedometers andpossible applied tovehicle trajectory control problem with aposition sensoraccelerometers that provide thevelocity states moredirectly andavelocity sensor. thantheindirectly observed values fromposition.