Chad Jenkins, Assistant Professor of Computer Science, has been awarded a three-year $380,000 grant from the Office of Naval Research. This grant, titled "Learning Predictive Motion Vocabularies for Kinematic Tracking and Activity Recognition", will support Jenkins' neuro-inspired research on learning the basic building blocks of human motion, or motion primitives, leading to their use for robot perception.
Important computational challenges exist in acquiring and analyzing large amounts of human motion. Existing work in this area has addressed some of these challenges for learning and vision-based tracking methodologies, but has been limited to relatively small datasets. Using new motion capture technologies, this funded project aims to greatly increase both the scalability of dimension reduction methods for large datasets and the breadth of computational priors for robot perception of human motion.
Jenkins' work builds on learning predictive models of human motion towards realizing a foundation for real-time kinematic tracking and activity recognition for human-robot interaction. Through data-driven development of predictive motion vocabularies, Jenkins seeks to enable new forms of human-robot interaction. Potential applications resulting include control of neural prosthetics, robot programming by demonstration, human-robot teams, remote monitoring, and robot programming for non-technical users.