Optimal Control of Camera Mounted Robots in Order to Maximize Information

Abstract

We consider the problem of optimal control of two sensory robots with on board cameras from the viewpoint of maximizing the accuracy of imaging of a set of moving targets. Both calibrated and uncalibrated camera systems are considered, and a performance index in the form of the integral of the error covariance of the reconstructed images, is suggested. In the uncalibrated case image reconstruction is only possible only up to the orbits of the projective linear group, and in this case we treat the covariance as a two tensor on the projective space. We also discuss the case of three dimensional camera motion under the Lising's constraints, and show how to extend planar results to this case.