A Feature-based Solution for Kidnapped Robot Problem

Abstract

The Kidnapped Robot problem in robotics commonly refers to a situation where an autonomous robot can’t locate itself against the map. It can be caused by external force when a robot is carried to an arbitrary place, or experiencing a wake-up sequence, etc. In this thesis, a solution is proposed and implemented using feature matching to solve the problem in an efficient and robust way. First a flood coverage algorithm is proposed to mark the building with limited number of fiducial markers and environmental features. Then a match-then-vote method is implemented based on ORB (Oriented FAST and Rotated BRIEF) and SIFT matching algorithms to recognize the place. The robot’s position can then be estimated by its distance and angle to the marker. After calibration by an AMCL (Adaptive Monte-Carlo Localization) node in ROS (Robot Operating System) using feature matching, the position message will be published in the robot operating system to locate the robot in a map. After tested in Auburn University Broun Hall, the solution proves to be a functional method in kidnapped recovery with acceptable error and speed: The average error of the distance measurement is below 1%. The average error in position estimation is about 0.31m while the angle error in the pose estimation is less than 3.5 degrees. And the recovery process with marker in sight cost less than 1.5 second.