Coordinating Team Tactics for Swarm-Vs.-Swarm Adversarial Games

Date: Wednesday, December 18, 2019

Time: 10:00 AM - 12:00 PM (EST)

Location: EBB Krone, Conference Room 4029

Laura Strickland

Robotics Ph.D. student

School of Interactive Computing

Georgia Institute of Technology

Committee:

Dr. Matthew Gombolay (Advisor) - School of Interactive Computing, Georgia Institute of Technology

Dr. Charles Pippin - Aerospace, Transportation and Advanced Systems Laboratory, Georgia Tech Research Institute

Dr. Seth Hutchinson - School of Interactive Computing, Georgia Institute of Technology

Dr. Magnus Egerstedt - School of Electrical and Computer Engineering, Georgia Institute of Technology

Dr. Frank Dellaert - School of Interactive Computing, Georgia Institute of Technology

Abstract:

The coordination scheme proposed in this work will enable large-scale teams of fixed-wing UAVs to maximize their utilities in zero-sum games. Each team consists of homogeneous units, while heterogeneity exists across teams. Many variables can influence the outcome of such a scenario, such as agent motion dynamics, team size, probability of disabling another agent, communication, and sensing. An effective team strategy is comprised of effective maneuvers for one-on-one scenarios, flexible coordination that leverages the team's offensive force in the changing environment, and accurate opponent assessment and prioritization.

To realize effective teaming, this work details team-tactical approaches to swarm-vs.-swarm games and proposes the development of a coordination schema for distributing a team's members such that the team’s utility is maximized throughout the game. Specifically, this thesis will make the following contributions: 1) Coordination analysis: Execute and analyze simulations of swarm-vs-swarm games to compare coordinated and non-coordinated tactics; 2) Tactical analysis: Examine relationships between the variables affecting such games to establish which are most impactful, and in what situations; and 3) Coordinated tactics: Develop a partitioning algorithm that combines search with the learning of policies for sections of the game space that maximizes the utility of the game space.