CIMMS Lunchtime Series: Bio-Inspired Visuomotor Convergence with Applications to Autonomous Flight Control and Navigation

Sean Humbert, Caltech, Mechanical Engineering

Prevalent in many biological control systems is the phenomenon of sensorimotor convergence, where signals from arrays of spatially distributed and differentially tuned sensors converge in number up to several orders of magnitude to motor neurons responsible for controlling locomotive behavior.

An excellent example occurs in the processing of retinal image pattern movement (optic flow) by the visuomotor systems of insects. Optic flow, composed of vantage point motion and the spatial distribution of objects in the environment, is a rich source of information for flight stabilization and navigation tasks. In insects, optic flow is parsed by wide-field motion sensitive interneurons (tangential cells, or LPTCs) in the lobula plate section of the visual ganglia, whose outputs synapse in the motor control centers of the animal.

This sensorimotor convergence technique, termed spatial wide-field integration, is used by insects to extract relevant information from the retinal motion patterns to modulate locomotion and flight. Within the control-oriented framework presented here, a theory of wide-field integration of retinal image flow is developed which establishes the connection between image flow kernels (retinal motion pattern sensitivities) and the output feedback terms they represent.

Through application to vehicles admitting euler-lagrange dynamics, it is shown that wide-field integration outputs can be used to shape the potential energy and inject damping, thereby stabilizing the different modes of planar flight and navigation. It is also demonstrated that the proposed output feedback methodology is sufficient to explain experimentally observed navigational heuristics as the centering and forward speed regulation responses exhibited by honeybees.