On Bayesian Inference for Embodied Perception of Object Poses

This extended abstract describes a probabilistic approach for improving the object pose estimation accuracy of embodied agents. We propose a histogram filter that uses (and supports) the exploration capabilities of robots, and allows for the integration of multiple methods. We discuss the design of experiments to estimate the error model of a pose detection method, and present a suitable representation of the orientation histograms. This allows us to consider priors about likely object poses or symmetries and use information gain measures for method and view selection. The method is validated and compared to alternatives using real-world 3D images acquired using a next-best-view algorithm