DrIan Abraham

Recent advances in design, control, learning have enabled gymnastic-level athleticism and extreme dexterity in robotic systems.  However, these robotic capabilities are facilitated by large-scale simulation and immense compute that do not necessarily scale to real-world settings that require on-line learning and adaptation. The underlying challenge is that existing methods are simply not computationally efficient, robust, or reliable to adapt to unforeseen settings, requiring retraining of robot behaviors via similar large-scale tactics. To truly advance the utility of robotic systems, we require careful study of how robots can robustly and selectively interact with and learn from their environment with minimal compute. Thus, my research program and interests seeks to analyze and discover computational methods that enable robots to curiously explore, interact, and learn in novel and remote scenarios with minimal data, compute, and with strong performance and reliability guarantees.