Autonomous Coordination Policies in Ground-Air Unmanned System Interaction
Abstract
Humans use a variety of complex strategies when solving problems. These strategies aresemantically ordered; that is, as the complexity of a problem increases, an explanation of astrategy used for a simpler problem would be related to that of a strategy used for a morecomplex problem. Machines, however, can learn complex strategies but they normallyproduce strategies that are different from those used by humans. The consequence is that it isunlikely that a human would trust the behaviour of, or coordinate activities properly with,these machines. The primary objective of this project is to design a machine learning algorithm that can learn incrementally while preserving the meanings embedded in a simpler context as they learn a more complex one. The algorithm will be tested in a simulation environment where a swarmof autonomous air vehicles coordinate activities with a swarm of land vehicles. This research will make a leap forward in the area of ground-air interaction, as well as in the field of reinforcement learning; in particular, transparent reinforcement learning where the control strategy can be explained and understood by a human subject. The success in achieving the aims of this project will have a significant impact on future heterogeneous swarm systems and swarm operations.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 09, 2018
- Source ID
- FA23861714054
Entities
People
- Hussein A. Abbass
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of New South Wales