High-speed visual robot control using an optimal linearizing intensity-based filtering approach

Many contact operations in robotics require accurate positioning, which is made difficult by the presence of rapidly varying interaction forces and compliances in gear boxes and links. In order to compensate for such effects, rapid feedback from the measured tool position in several degrees of freedom is needed. This paper presents a dynamic visual tracking technique based directly on intensity measurements in the image, which can be used to obtain state estimates at a very high rate, and with very short input-output latency. Methods for analysis of the stability and sensitivity to disturbances are presented, and an improved version for better disturbance suppression of illumination variations and noise is developed. Positioning experiments using an industrial robot with camera feedback at 250 Hz are used to validate the approach