Garrett Katz and Di-Wei Huang Win Best Student Paper at AGI 2016

For their work entitled "Imitation Learning as Cause-Effect Reasoning," PhD students Garrett Katz and Di-Wei Huang won Best Student Paper at the Ninth Annual Conference for Artificial General Intelligence (AGI 2016) held in New York, New York from July 16-19, 2016.  Professor Jim Reggia advises both students. Katz and Huang have also collaborated with Professor Rodolphe Gentili of the Kinesiology Department.

(L-R: Katz, Huang)

Katz explained what comprises imitation learning in robots. 

"The robot learns new skills by observing a human demonstrator rather than being programmed by a robotics expert," he said. "Unlike other approaches, our robot uses cause-effect reasoning to form a deeper understanding of why the demonstrator did what they did.  This lets the robot successfully carry out the same skill in different situations that require different actions."

Huang provided videos that demonstrate what robots (specifically Baxter from Rethink Robots) are able to learn from a single example move. 

In his research, Katz, a fourth-year PhD student, focuses on neurocomputational cause-effect reasoning in the context of robotic imitation learning.  Huang, who expects to graduate in August of 2016, focuses on neural-based machine learning.  His dissertation research focuses on self-organizing map neural architectures based on limit cycle dynamics. Haung also works on an ONR project studying effective imitation learning for robots.

In an enthusiastic correspondence about his students' work, Professor Jim Reggia said, "Garrett and Di-Wei have developed a new methodology that allows a person to demonstrate a simple task to a robot in a virtual world, and then the robot can learn from this demonstration to imitate that task in the real world." He continued, "Importantly, the robot uses cause-effect reasoning to try to understand why the person is doing what it's seeing, rather than trying to just literally copy the person's actions. Not only is this cost effective, but it also allows a much broader range of people than in the past to program robots to carry out tasks."

Katz, G., Huang, D. W., Gentili, R., & Reggia, J. (2016, July). Imitation Learning as Cause-Effect Reasoning. In International Conference on Artificial General Intelligence (pp. 64-73). Springer International Publishing.

Preprint available.