Adaptive Kalman Filter Control Law for Visual Servoing

Abstract

We improve uncalibrated visual servoing for multiple cameras. This moves robots towards being more effective in unstructured environments. Our approach improves performance and reliability by prioritizing data from particular cameras in the system. Adaptive filtering is used to effect this in an uncalibrated situation. Rules are introduced for ensuring positive definite covariance estimates, with the resulting control law then compared to two prior methods. Performance metrics show that the new control law yields lower tracking error than previous approaches and the standard deviations of these metrics indicate that it provides a much more reliable system. Simulation results include improvements up to 33% in the performance metric for a static target and 63% in standard deviation of the performance metric for a moving target.