Dr. John Fossaceca of Army Research Laboratory

In the future, it is likely that the Army must be prepared to engage in distributed operations in complex environments under extreme resource constraints with novel or limited advance training data. It will not be possible to plan for every possible event nor will it be possible to collect enough data to train deep learning systems to handle every possible situation. Yet the need for autonomous systems that can reason about and act in complex environments is essential for successful execution of Multi-domain Operations (MDO). This talk begins with a brief introduction to the AI for Maneuver and Mobility Essential Research Program at the Army Research Laboratory and some of the Army’s unique technical challenges that the commercial sector will not address. Next we will cover some of the promising research on human-robot interaction and discuss ARL’s “Cycle-of-Learning for Autonomous Systems from Human Interaction” including learning from intervention (LFI), learning from evaluation (LFE) and learning from demonstration (LFD) with some specific examples of the new methods. This research aims to provide the ability for autonomous vehicles and other intelligent agents to learn on-the-fly and operate effectively in novel and complex environments by learning directly from Soldiers in the field. Recent research on learning, specifically robots learning from humans has shown great promise, drastically reducing the training time required for reinforcement learning, avoiding manual parameter re-tuning in new environments and decreasing the need to collect enormous amounts of a priori training data.

Biography:

John M. Fossaceca, Ph.D., is the Program Manager (A) for the Artificial Intelligence for Maneuver and Mobility Essential Research Program at the U.S. Combat Capabilities Development Command, Army Research Laboratory. The AIMM-ERP seeks to perform foundational research to support AI-enabled systems for autonomous maneuver that can rapidly learn, adapt, reason, and act in multi-domain operations. Earlier in his career, Dr. Fossaceca was VP of Engineering and Product management at 3e Technologies International (3eTI), a division of Ultra Electronics, where he was principal investigator on several research programs and was instrumental in creating secure wireless systems for the U.S. Navy. As Engineering Vice President and COO at Comtech Mobile Datacom he led modernization effortsfor the U.S. Army's Blue Force Tracking command and control satellite-based tracking system. Dr. Fossaceca previously served as Technical Director for AT&T/Lucent Technologies/Bell Labs in both Holmdel and Murray Hill, NJ contributing to wireless telephony products and next-generation VoIP and packet-based communication systems. He has published research on machine learning for network intrusion detection and is co-inventor on six patents related to wireless communications, and adaptive systems. Dr. Fossaceca is an adjunct professor at George Washington University teaching graduate courses in Machine Learning and Cybersecurity and holds a Bachelor's degree in Electrical Engineering from Manhattan College, a Master's degree in Electrical Engineering from Syracuse University, and a Ph.D. in Systems Engineering from George Washington University.