Zoom Link: https://upenn.zoom.us/j/98563756560?pwd=yXnuhaTv2V2jSOHzgiz1DujOa2Ubdw.1

Robots still have a lot to learn from animals when it comes to agile locomotion. Abstracting low-DOF dynamical models from animal locomotion and using them as control targets for robots has demonstrated impressive empirical results while often facilitating formal performance guarantees. However, implementations of this approach have historically tended to restrict their focus to sagittal-plane dynamic models, and thus have been implemented primarily on robots with few degrees of freedom. These limitations have largely prevented the associated biologically-inspired behaviors from finding application in the real world.

This thesis presents several contributions that bring this compositional biologically-inspired approach out of the sagittal plane and into three-dimensional space. First, it presents a framework for anchoring sagittal-plane templates in spatial quadrupeds, using their out-of-plane actuated degrees of freedom to stabilize the motion within the plane, and thus demonstrating how such modular behaviors can find application across embodiments. Further analysis of the orientation controller used in this anchoring reveals a broader class of “pointing direction” controllers that provide exponential stability within an explicitly characterized basin of attraction within the tangent space of spatial rotations. This class of controllers is sufficiently general to extend to other applications, including quadrotor and satellite control. Next, the thesis presents a novel composition of three-dimensional templates that generates a fully spatial bounding gait, and provides experimental evidence of robustness sufficient for locomotion over natural terrain. Finally, it introduces a further extension of the orientation controller that significantly improves performance in real world settings, providing both formal proofs of local stability and experimental evaluation of its effective basin of attraction. These contributions illustrate the power of such biologically-inspired compositional approaches to generate agile, platform-independent behaviors that admit formal stability guarantees.