Multiple Hypotheses
	* Represent a multi-modal    posterior probability    distribution over model    parameters      - sampled representation      - each sample is a pose      - predict over time        using a particle        filtering approach.