Optimizing Interaction Potentials for Multi-Agent Surveillance
Abstract
We have developed a physics-based control framework that provides a practical yet principled approach for designing collective systems. This framework is called "artificial physics" (AP), because agents perform actions based on virtual forces exerted on them by other agents and the environment. These forces are designed to ensure that the global behavior of a multi-agent system arises from local interactions of the agents, as well as from task-specific goals and constraints. We extend AP by using genetic algorithms (GAs) to search a space of interaction potentials so that the desired behavior emerges from the interactions between the agents. This extended framework is applied to the task of surveillance, where a team of unmanned air vehicles (UAVs) must provide maximum sensory coverage of terrain, in order to maximize the probability of detection of targets of interest. This report summarizes preliminary results that indicate that robust behavior is achieved, despite loss of assets or sensor degradation. This report also provides some initial theoretical analyses of simple behavior-based asset controllers on the surveillance task.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2004
- Accession Number
- ADA434929
Entities
People
- Diana Spears
- Dimitri Zarzhitsky
- Suranga Hettiarachchi
- Wesley Kerr
- William Spears
Organizations
- University of Wyoming