Abstract:
Many robots can be categorized into similarity classes like robot arms or snake robots. Despite their kinematic differences, we can intuitively recognize that two different robot arms often perform visually similar motions. However, their joint space representations do not reflect our intuitive notion of visual similarity. We believe that there exists an abstract shape space that can better capture similarities in robot motions and present methods for finding and interacting with that shape space. Furthermore, by interpreting the shape space as a manifold, there is rich differential geometry theory to address the problem of projecting trajectories across different robots. In this talk, we explore shape spaces for robot arms and kinematic snake robots. We show through experiment how one trajectory in the unified robot arm shape space can be executed on different arm models with different degrees of freedom. Next, we show for kinematic snakes how the framework of unified shape spaces also allows for the execution of a velocity-space trajectory across different snakes.

Committee:
Howie Choset (Chair)
Jeff Schneider
Wennie Tabib
Ben Freed