The skeptical explorer: A multiple-hypothesis approach to visual modeling and exploration

The primary intent of this work is to present a method for sequentially ass ociating three-dimensional surface measurements acquired by an autonomous e xploration agent with models that describe those surfaces. Traditional mult iple-viewpoint registration approaches are concerned only with finding the transformation that maps data points to a chosen global frame. Given a part s-based object representation, and assuming that the view correspondence ca n be found, the problem of associating the registered data with the correct part models still needs to be solved. While traditional approaches are con tent to group segmented data sets that geometrically overlap one another wi th the same part, there are cases where this causes ambiguous situations. T his paper addresses the model-data association problem as it applies to thr ee-dimensional dynamic object modeling. By tracking the state of part model s across subsequent views, we wish to identify possible events that explain model-data association ambiguities and represent them in a Bayesian framew ork. The model-data association problem is therefore relaxed to allow multi ple interpretations of the object's structure, each being assigned a probab ility. Rather than making a decision at every iteration about an ambiguous mapping, we look to the future for the information needed to disambiguate i t. Experimental results are presented to illustrate the effectiveness of th e approach.