Structure-Independent Visual Motion Control of the Essential Manifold

Abstract

A new framework for visual motion control is described, which consists of formulating the control task on the so-called essential manifold, a "compact" matrix representation of SE(3). Unlike previous image plane control techniques, our method does not require information about the geometric structure (depth) of the scene or target. This allows us to design control laws that are not ill-conditioned when close to zero-disparity configurations. The control relies on a causal motion estimator (called the "essential filter") that identifies recursively an implicit dynamical model with parameters on the essential manifold.

Files

Additional details