Neuromorphic Processing of Moving Sonar Data for Estimating Passing Range

Ranging sensors typically estimate range in order to register object locations with respect to a floor plan. This paper uses range readings from a moving sonar to estimate the passing range, equal to the minimum range as the sonar passes by the object. Estimating passing range not only indicates if a collision will occur, but also leads to artifact rejection and object classification. We control a conventional sonar to generate a spike process whose density relates to echo waveform intensity, analogous to biological action potentials. While the sonar moves along a linear trajectory, it extracts strong echoes and stores their range measurements in memory. Neuromorphic processing applies delays and coincidence detection to passing-range estimates for localizing and classifying objects. Physical principles governing echo production motivate a multiresolution coincidence detector that accomplishes the important sensing tasks of object classification, collision avoidance, trajectory alignment, artifact rejection, and sonar data fusion. Objects are classified by their hyperbolic range readings that exhibit passing-range estimate coincidence at a resolution related to surface roughness. Distributed objects parallel to the sonar trajectory, such as rough surfaces, exhibit coincidence in range readings. Multiple coincidence detectors tuned to different step sizes can adapt to changes in translational speed by changing the probing pulse period. Drive-by sonar experiments past an isolated post and down a hallway containing retroreflectors and distributed reflectors indicate localization and classification capabilities of passing-range estimation