Intelligent sensing and machine reasoning for robotic manipulation of irregular moving objects

The authors describe research in areas related to the automation of package handling in the US Postal Service, particularly the acquisition and interpretation of range images, and experiments with a robotic singulator. It is noted that these projects considered but did not lead to immediate applications in the postal environment. Several conjectures about what is needed to create a breakthrough are described. In machine vision, it seems important to develop techniques to selectively image the surroundings and extract information as needed to update a world-view, and to think in terms of streams of images quickly processed instead of exhaustive processing of individual images. Basic sensing capability still presents technical challenges, but sensing and interpretation of signals from multiple sensors should be developed together. It is suggested that the entire reasoning and control structure can be advantageously constructed as a hierarchy