Surprisingly, a thorough experimental evaluation of the accuracy related properties of differential visual feedback control has never, to our knowledge, been performed. The closest approach is [Wije et al 93], who showed for a 2 DOF implementation, the advantage of visual feedback over open loop control when model errors are large. Chen et al in [Chen et al 94] show that two heavy industrial robot arms can perform a peg-in-hole parts mating task with a clearance of 0.7mm between the peg and the hole.
In this section we describe some of the characteristics of our adaptive control method that makes it particularly suited to the kind of visual space specification and control we do. First, we evaluate repeatability under visual feedback control, and compare the results to standard joint level control. Then we demonstrate how a 3 DOF adaptive controller is significantly better at dealing with the typically large model errors of a uncalibrated vision system than a fixed gain controller. A 6 DOF controller is generally even more severely affected by model errors. In a second experiment, with a 6 DOF system, we show that as long we stay within reasonable error bounds in our linear model we typically achieve convergence to 0 pixel error in image space. Finally we motivate the importance of using redundant perceptual information by tracking many more features than the degrees of freedom under control, showing that absolute (world) positioning accuracy improves for a visual processing front end with fixed accuracy on tracking individual features.