SUMMARY

This thesis presents a mathematical framework and methods to coordinate motion control and image processing in a fast moving robotic system. The human eye has two representative movements, a saccade and smooth pursuit. Saccade is one of the fastest and most accurate movements in the human eye. However, the human visual system receives blurry environmental information due to a finite integration time when a saccade takes place. Smooth pursuit is another eye movement that continuously follows an object with relatively slow velocity. This thesis presents dynamics-based image processing methods for a fast-moving robotic camera system, inspired by the observation of the physiological evidences. Real-time panoramic image stitching is presented with simultaneous motion de-blurring in a dynamic vision system, allowing for generic image sensors with a standard frame rate and significantly less computational load, and requiring no motion sensors. The proposed methods are based on the dynamic model of the robotic eye and movements induced by the control system while conventional methods analyze inherent image properties. In the neuromotor system, the movements are generated by muscles that are essentially quantized, compliant actuators. To reproduce smooth pursuit-like movements, an open-loop discrete switching controller accounts for dynamics is introduced to generate an arbitrary velocity profiles for a robotic eye system driven by quantized, compliant actuators, while avoiding high-frequency switching in individual motor units. The effectiveness of the dynamics-based methods and the discrete switching controller are presented and validated on a fast-moving robotic orientation system.