Abstract
Robots are becoming increasingly available at both public and private places, and the amount of videos from them is explosively rising. In order to provide activity-level situation awareness to robots and make them operate naturally in its environment while interacting and collaborating with other humans/robots, algorithms to understand such robot-centric videos are essential. In this talk, we introduce deep learning (i.e., convolutional neural network) models for video data and discuss how we can take advantage of them for activity recognition and robot action learning.   The objective is to learn activity models from human videos for both their recognition and robot imitation. We talk about activity recognition models with the concept of sub-events and super-events, and describe fully convolutional models for robot imitation learning.

Bio
I am an Assistant Professor in the Department of Computer Science at Indiana University. I also have an adjunct position at the NASA's Jet Propulsion Laboratory (JPL), California Institute of Technology as their Research Affiliate. Before joining IU, I was a Research Technologist within the Robotics Section of JPL from October 2011 to July 2015. Prior to that, I completed my national duty at ETRI, Korea, working as a research scientist. I received my Ph.D. from the University of Texas at Austin in 2008, and the B.S. degree from the Korea Advanced Institute of Science and Technology (KAIST) in 2004.