Abstract
This paper describes a method of visionโbased selfโvelocity estimation and its feedback system in force/torque sensorless power assisting control of a wheelchair robot. In this method, threeโdimensional information obtained from stereo images is used along with optical flow vectors for selfโvelocity estimation in real time. The human force is also estimated by a sensorless reaction force observer, and the assisting force is calculated by using an estimated force and virtual impedance model. In the paper, the forceโbased assist function is integrated into a visual feedback motion controller. This approach using visionโ and forceโbased assist control makes it possible to facilitate direct intelligent interactions between human force and environments such as human following assistance, obstacle avoidance assistance, and so on. Such assistance functions are changeable by selection of the weighting matrix used in velocity estimation, which is based on weighted least square solutions obtained from optical flow vectors. The validity of the proposed approach is verified by several experimental results. ยฉ 2009 Wiley Periodicals, Inc. Electr Eng Jpn, 169(2): 21โ28, 2009; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/eej.20887