Sensorimotor Convergence in Natural Systems and Applications for Autonomous Microsystems

Dr. J. Sean Humbert
Department of Aerospace Engineering
University of Maryland, College Park

Natural systems, particularly smaller animals with significant processing and energy constraints, have developed rapid and effective methods of extracting information from their environments. The observed architecture consists of distributed and differentially tuned arrays of simple sensors whose spatiotemporal patterns are decomposed with static weighting (sensitivity) functions. It is remarkable that the resulting outputs are often communicated directly to motor neurons, without the intermediate estimation stage that is prevalent in their artificial counterparts. This suggests efficient sensor-actuator pairings have evolved to extract only the critical information required for successful behavior. In this talk the advantages of this unique sensorimotor architecture will be discussed, with an emphasis on application to microsystem autonomy and mobility. Examples will be drawn from visual (optic flow) and mechanosensory (hair deflection) sensory patterns. It is shown that relevant information for navigation and flight control can be extracted with minimal latency and computation power, providing a solution to the processing throughput problem that is consistent with the payload and closed-loop bandwidth constraints of small MAVs.