Articulo robots

Journal of Bionic Engineering 7 Suppl. (2010) S172–S176

Self-Localization of an Omnidirectional Mobile Robot Based on an Optical Flow Sensor
Atsushi Sanada1, Kazuo Ishii1, Tetsuya Yagi2
1. Kyushu Institute of Technology, 2-4 Hibikino, Wakamatsu, Kitakyushu, Fukuoka 808-0196, Japan 2. Osaka University, 2-1 Yamadaoka, Suita, Osaka 565-0871, Japan

Abstract
An omnidirectional mobile robothas the advantage that three degrees of freedom of motion in a 2D plane can be set independently, and it can thus move in arbitrary directions while maintaining the same heading. Dead reckoning is often used for self-localization using onboard sensors in omnidirectional robots, by means of measuring wheel velocities from motor encoder data, as well as in car-like robots. However, omnidirectionalmobile robots can easily slip because of the nature of omni-wheels with multiple free rollers, and dead reckoning will not work if even one wheel is not attached to the ground. An odometry method where the data is not affected by wheel slip must be introduced to acquire high quality self-location data for omnidirectional mobile robots. We describe a method to obtain robot ego-motion using cameraimages and optical flow calculation, i.e., where the camera is used as a velocity sensor. In this paper, a silicon retina vision camera is introduced as a mobile robot sensor, which has a good dynamic range under various lighting conditions. A Field-Programmable Gate Array (FPGA) optical flow circuit for the silicon retina is also developed to measure ego-motion of the mobile robot. The developedoptical flow calculation system is introduced into a small omnidirectional mobile robot and evaluation experiments for the mobile robot ego-motion are carried out. In the experiments, the accuracy of self-location by the dead reckoning and optical flow methods are evaluated by comparison using motion capture. The results show that the correct position is obtained by the optical flow sensor rather thanby dead reckoning. Keywords: omnidirectional mobile robot, silicon retina camera, optical flow, FPGA
Copyright © 2010, Jilin University. Published by Elsevier Limited and Science Press. All rights reserved. doi: 10.1016/S1672-6529(09)60232-8

1 Introduction
One of the key words in current mobile robot research is "omni", as in omni-vision and omnidirectional motion. In the RoboCup Middle SizeLeague (MSL), most of the robots have omnidirectional vision and omni-wheels, which is required for optimization under RoboCup MSL rules[1–3]. The omnidirectional mobile robot has the advantage that three degrees of freedom of motion in a 2D plane can be set independently, and it can thus move in an arbitrary direction while maintaining the same heading. Dead reckoning is often used forself-localization using the onboard sensors in the omnidirectional robot, by means of measuring the wheel velocities from motor encoder data, as well as in car-like robots. However, omnidirectional mobile robots can easily
Corresponding author: Atsushi Sanada E-mail: sanada-atsushi@edu.brain.kyutech.ac.jp

slip because of the characteristics of omni-wheels with multiple free rollers, and dead reckoningwill not work well even if one wheel is not attached to the ground. An odometry method where the data is not affected by wheel slip must be introduced to acquire high quality self-location data for omnidirectional mobile robots. We describe a method to obtain the position of a robot using camera images and optical flow calculation[4,5], i.e., where the camera is used as a velocity sensor. Opticalflow is obtained as the difference between successive images, but it requires considerable computational power to realize real-time operation, while the system must be small and have sufficiently low energy consumption to be mounted on the mobile robot. In this paper, a silicon retina vision camera developed by Yagi et al. is introduced as a mobile robot sensor, with a good dynamic range under...