MORE | Fall 2023

Deep Learning Based Changepoint Detection for Robot Learning

The research team is developing an algorithmic approach for changepoint detection to segment complex, long-horizon tasks into smaller, more manageable subtasks for robot learning. By utilizing statistical testing methods and video encoding models, we aim to reduce computational complexity and enable efficient learning in high-dimensional robotics domains. The anticipated outcomes include precise task segmentation, successful transfer of learning to reinforcement phases, and seamless multi-task learning for robots. Beyond robotics, this approach holds potential for applications in healthcare and other industries where changepoint detection is valuable for data analysis and decision-making.