Vision-Based Obstacle Avoidance of Wheeled Robots Using Fast Estimation

摘要

THE application of mobile vehicles for military and commercialnpurposes has increased dramatically in recent years. To minimizenhuman interaction, these vehicles must be able to execute robustnand efficient obstacle detection and avoidance using onboard measurements.nInertial measurement units augmented with GPS haventraditionally been used for onboard sensing in a wide variety of missions.nThe reliance of GPS on low-power radio signals from Earthorbitingnsatellites makes it vulnerable to intentional (e.g., jamming)nor unintentional (e.g., electrical) interference that can degrade systemnperformance [1]. Alternatively, vision-based sensors are being usednas they are low cost, light weight, and passive. Vision-based sensorsnrequire a robust estimation scheme. The extended Kalman filtern(EKF) has been used widely for this purpose [2–7]. Application ofnthe EKFrequires linearization about the desired trajectory and is verynsensitive to initial errors [8–10]. Improvements of EKF performancenwith application to obstacle avoidance have been reported in [8,9,11]nby using unscented Kalman filters and sigma-point Kalman filters.nHowever, convergence guarantees for the parameter and range estimationncannot be deduced from application of the EKF or itsnvariations.