Getting The Big Idea Recognizing Plans And Activities By Reading People Like A Book
Abstract: It has been suggested that plan recognition (PR) and natural language processing (NLP) have much in common and are amenable to similar analyses. PR techniques often focus on the structural relationships between consecutive observations and ordered activities that comprise plans. However, NLP sometimes treats text as a bag-of-words, omitting such structural relationships and using topic models to get the overall distribution of concepts discussed in documents. We examine an analogous treatment of plans as distributions of activities through the application of Latent Dirichlet Allocation topic models to human skeletal data of plan execution traces obtained from a RGB-D sensor. This talk focuses on how to represent the sensor data as text and explores early work into how LDA performs integrated PR and activity recognition. We will also introduce variants of LDA for lifted recognition and propose future variants that will be practical for general-purpose applications including human-robot interaction.
Bio: Rick Freedman is a second-year MS/Ph.D. student in the School of Computer Science at the University of Massachusetts Amherst. He received his BS in computer science and mathematics from Wake Forest University. As a member of Dr. Shlomo Zilberstein's Resource Bounded Reasoning Lab, his research interests in artificial intelligence include plan recognition and generalized planning. He participates in interdisciplinary research and often integrates these areas with research in statistical-relational artificial intelligence, human-robot interaction, and natural language.
Additional Information for Attendees: This work is a collaboration between the Resource Bounded Reasoning (RBR) Lab, Laboratory for Perceptual Robotics (LPR), and Machine Learning and Data Sciences (MLDS) Lab at Universtiy of Massachusetts Amherst. It is in progress, and feedback will be appreciated. The presentation is aimed at a general audience.