Over the past decade, machine learning has enabled automated decision-making algorithms with superhuman proficiency in challenging domains, including the game of Go and robotics. However, it remains an open challenge to bring the power of automated decision-making back into the machine learning pipeline—particularly for data collection. My research develops robot autonomy for unsupervised data collection, with the goal of creating an “AlphaGo” moment in how machines acquire data, achieving superhuman proficiency with certifiable optimality. In this talk, I will highlight three areas of my research that are crucial to this goal: (1) control for physical randomization, which enables robots to generate data with certifiable statistical properties through motor control; (2) heterogeneous multi-agent coordination, which leverages the complementary strengths of different agents, sensors or actuators; and (3) hierarchical decision-making, which exploits hierarchical structure to augment continuous control with discrete search.
Max Muchen Sun
Max Sun is a final-year Ph.D. student at Northwestern University’s Center for Robotics and Biosystems, advised by Prof. Todd Murphey. His research focuses on developing robot autonomy as controllable data engines that generate data without human supervision. Applications of his research include robot manipulation, robot perception, human–robot interaction, and soft robot control. He also collaborated with Dr. Peter Trautman at Honda Research Institute, where they developed a full-stack system for a large-scale field deployment of robot navigation in human crowds in Santa Cruz, CA.
This event is open to NYU students, faculty, and staff.
📍 In-Person Location: 5 MetroTech LC400 [NYU ID required]
📍 Online Access: Luma
📍 Meeting ID: TBD
📍 Meeting Passcode: TBD