PhD Thesis Defense

Abstract: Data-driven learning is a powerful paradigm for enabling robots to learn skills. Current prominent approaches involve collecting large datasets of robot behavior via teleoperation or simulation, to then train policies. For these policies to generalize to diverse tasks and scenes, there is a large burden placed on constructing a rich initial dataset, which is [...]

Abstract:   Current trends in robotics design and engineering are typically focused on high value applications where high performance, precision, and robustness take precedence over cost, accessibility, and environmental impact.  In this paradigm, the capability landscape of robotics is largely shaped by access to capital and the promise of economic return. This thesis explores an alternative [...]