My talk will focus on theoretical and algorithmic ideas in machine learning and their origin in problems of robotics. Much of my talk will focus on no-regret online learning methods in machine learning and the critical role of interaction for learning in robotics.
I will highlight the tremendous impact of robotics in identifying key learning problems and in suggesting algorithmic techniques; conversely, I'll consider the remarkable tools that have been developed within learning to address hard robotics problems. I'll discuss a spectrum of machine learning techniques of increasing sophistication from the most familiar classification problems, to structured prediction, to imitation learning, to making reinforcement learning and learning control practical in robotics.
Throughout, we will look at case studies in learning dexterous manipulation, activity forecasting of drivers and pedestrians, to imitation learning of robotic locomotion and rough-terrain navigation. These case-studies highlight key challenges in applying learning algorithms in practical settings.
sharonw [atsymbol] cs.cmu.edu