Optimal collaborative multi-agent disruption theory with antagonistic agents
Abstract
One of the main current problems of interest to the defense community is that of dealing with very largenumbers of aggressive collab,orative agents. In general, when the number of agents in an attacking system are relatively low, the current approach is to destroy, the agents with kinetic or high-power electromagnetic weapons. However, in the future, collaborative multi-agent adversaries will b,e communicating and reactive with built-in autonomy, and will consist of very large numbers as well. In order to defend Navy assets, from such threats, it will be important to use other autonomy-based approaches that discern the opposingswarm s dynamics and access,ible design and control parameters. If one can estimate a swarm s future dynamics, how it reacts to stimuli such as other agents and, its environment, then one has a better chance of disturbing its current dynamics and possibly redirecting its target trajectory to, one of less importance.The objective of the proposal is to design mechanisms for actively probing an opposing multi-agent system s, state, predicting its spatiotemporal dynamics, disrupting its dynamic pattern formation, and modifying its behavior using antagonis,tic agents. Specifically, we will address issues o f: using a small multi-agent team as a probe to discover a larger adversary s dyn,amics, deriving and testing the numbers of agents needed to observe measurable changes in the behavior of an adversarial multi-agent, system or swarm, modifying the swarm optimally with a minimum number of injected antagonistic agents, andfinding the optimal " knob,s" to change/direct its behavior.Our approach will be to consider multi-agent collaborative systems and swarms that are reactive, su,ch as those formations that have collision sensing built into the agents. We will then take the followingapproaches to predict the s,warm s dynamics by: 1) Determining how many probing or antagonistic agents it takes to detect a measurable response in the opposing, swarm; 2) Determine what perturbation patterns of controllable antagonistic agents will change the trajectory or split the swarm; a,nd 3) Identify accessible parameters using bifurcation theory and physics-informed machine learning to predict the opposing swarm s, objective and/or modify its target trajectory or dynamic pattern to one that is accessible.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 08, 2022
- Source ID
- N000142212157
Entities
People
- Mong-ying Hsieh
Organizations
- Office of Naval Research
- United States Navy
- University of Pennsylvania