Automatic verification of principal contacts using vision

Automatic recognition of unknown contacts among objects provides high-level feedback for a reactive robot system. The recognition mechanism must be online and insensitive to sensing uncertainties. A method is introduced of online using vision to extract more accurate contact information by reducing a set of initial contact hypotheses which are obtained from position/orientation sensing and subject to sensing uncertainties. It is assumed that contacts are among polyhedral objects with known geometric models. Preliminary implementation of the algorithms is discussed with testing results shown