Detection and Learning of Unexpected Behaviors of Systems of Dynamical Systems by Using the Q^2 Abstractions
Abstract
This research effort will focus specifically on swarms of UAVs performing various missions, like detecting and tracking all objects of interest in a region, detecting interesting activities at a variety of facilities by using dense in time still imagery from multiple UAVs. This effort will survey the potential missions that swarms of UAVs can carry out and represent the behaviors in ontology of undesirable behaviors. The ontology will include a classification of undesirable behaviors, their characteristics, as well as the relationships among the behaviors and the characteristics. The characteristics will serve as the basis for establishing the space of the variables that need to be monitored by the ground station or communicated among the agents. The outcomes of the research will include an ontology of undesirable characteristics of behaviors of multi-agent systems with respect to the missions related to information collection, object detection, and tracking; simulations exemplifying undesirable behaviors; global control policies that result in the emergence of different types of behaviors; algorithms for learning critical hypersurfaces for partitioning the system spaces into qualitative inputs, states and outputs and for constructing qualitative state machines; demonstration code; and results of formal analysis of the complexity and efficiency approach.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 11, 2016
- Source ID
- FA87501510095
Entities
People
- Mitch Kokar
Organizations
- Northeastern University
- Rome Laboratory
- United States Air Force