Co-Design of Bodies and Strategies

Pervan, Ana

doi:https://doi.org/10.21985/n2-80he-5e20

Work

Co-Design of Bodies and Strategies

Public

Download PDF

Download All Files (.zip)

The field of robot design mostly focuses on careful construction of complex control and planning algorithms (e.g., tuning neural network weights) which bear sole responsibility for improving task performance, while the robot’s body is often assumed to be part of the environment. In nature, however, biological organisms co-evolve both their neurological capabilities and their physical morphology to improve their chances for survival. This enables information and intelligence to be encoded not only in a centralized brain, but also in a distributed body. This thesis focuses on co-design of control algorithms and physical robot bodies. This thesis begins by introducing elements of robot design on a minimal, micro-robotic system. I analyze tasks (like micro-manipulation and target localization) in terms of the fundamental capabilities and information required to achieve them, and use those to generate robot designs ideally consisting of only the most essential components necessary for the task. The resulting designs are compared in terms of task performance and design complexity. These principles are then applied to the design of a group of robots, in which the collective system demonstrates emergent behaviors that the individuals are incapable of. Designing for ensemble behaviors requires predictions and analysis of inter-robot communication and collaboration. I compare the results of the collaborative robots with individuals attempting the same task to show that the emergent behaviors greatly benefit the system. I conclude this thesis by outlining a macroscopic extension of the contributions in which a robot co-designs a flexible tool by bending it into a shape that both informs and is informed by the control algorithm defining how the tool will be used to achieve a goal.

Creator