A Continuous-Time Nonlinear Observer for Estimating Structure from Motion from Omnidirectional Optic Flow

Abstract

Various insect species utilize certain types of self-motion to perceive structure in their local environment, a process known as <italic>active vision</italic>. This dissertation presents the development of a continuous-time formulated observer for estimating structure from motion that emulates the biological phenomenon of active vision. In an attempt to emulate the wide-field of view of compound eyes and neurophysiology of insects, the observer utilizes an omni-directional optic flow field. Exponential stability of the observer is assured provided the persistency of excitation condition is met. Persistency of excitation is assured by altering the direction of motion sufficiently quickly. An equal convergence rate on the entire viewable area can be achieved by executing certain prototypical maneuvers. Practical implementation of the observer is accomplished both in simulation and via an actual flying quadrotor testbed vehicle. Furthermore, this dissertation presents the vehicular implementation of a complimentary navigation methodology known as <italic>wide-field integration</italic> of the optic flow field.

The implementation of the developed insect-inspired navigation methodologies on physical testbed vehicles utilized in this research required the development of many subsystems that comprise a control and navigation suite, including avionics development and state sensing, model development via system identification, feedback controller design, and state estimation strategies. These requisite subsystems and their development are discussed.