Abstract
Traditional autonomy approaches demand high size, weight, and power sensors and computers for acquiring and interpreting rich sensory information to build a global rep resentation of the environment. This dissertation aims to address the research problem of lightweight aerial autonomy through the notion of lean sensing and embodied per ception. First, a comprehensive review studied state-of-the-art aerial robot design and codified existing design information into formal, reusable design principles. Fourteen design principles were derived, and five are relevant to achieving lightweight auton omy. Then, three technical developments are presented. These work demonstrates that lean sensing opportunities arises from the robot’s bodily interaction within its situated environment, i.e. embodied perception. They cover applications in standard quadrotor type and unconventional nature-inspired aerial robots. The first work focuses on spatial positional estimation for exteroception. The research utilizes contextual information of the environment as perceptual affordance, and design optimization of the robot’s sensor placement to improve sensing affordance. The robot’s desired trajectory and mechani cal constraints were incorporated to inform the optimization. The following work cen tres on proprioception: robot’s self angular rate estimation. The unique magnetometer waveform arising from the robot’s locomotion in an environment’s quasi-static electro magnetic field is incorporated to estimate its angular rate. Specifically, the frequency of the periodic signal generated is used to determine its rotation rate. The last work builds on the cumulative knowledge gained on the previous research, proposing a novel sensor-robot system for self-attitude estimation, spatial position estimation, and map ping applications. The robot’s underactuated and rotating locomotion are used to afford omnidirectional laser scanning of the environment with a single unidirectional laser. The prominent features of a situated indoor environment are incorporated to simplify the al gorithm complexity and computational power. Finally, key contributions of this disser tation are summarized, and recommendation for future work are discussed.