Hierarchical information combination process for large-scale task and team environments

In this paper, we describe the Distributed Dispatcher Manager (DDM), a system for monitoring large collections of dynamically changing tasks. We assume that tasks are distributed over a virtual space. Teams consist of very large groups of cooperative mobile agents. Each agent has direct access to only local and partial information about its immediate surroundings. DDM organizes teams hierarchically and addresses two important issues that are prerequisites for success in such domains: (i) how agents should process local information to provide a partial solution to nearby tasks, and (ii) how partial solutions should be integrated into a global solution. We conducted a large number of experiments in simulation and demonstrated the advantages of the DDM over other architectures in terms of accuracy and reduced inter-agent communication. This paper considers the problem of monitoring large collections of dynamically changing tasks. The tasks are distributed over a large (possibly, virtual) environment and are to be executed by large teams of mobile cooperative agents. These agents have direct access to only local and partial information about their immediate environment. There are several domains where such problems arise: satellites that are tasked to form a general picture of a large area; satellites that form weather maps; agents that control air pollution or ocean pollution; sensor webs that monitor geographic areas for passing aircrafts; and unmanned air and ground vehicles that must be jointly tasked for surveillance missions. In such domains, there are two central issues that represent prerequisites for success: (i) how agents should process local information to provide a partial solution to nearby tasks, and (ii) how partial solutions should be integrated into a global solution. We describe the Distributed Dispatcher Model (DDM), an agent based computational model. DDM is designed for efficient coordinated task allocation in systems consisting of hundreds of agents (resources); the model makes use of hierarchical group formation to restrict the degree of communication between agents. Our main contribution is in use of a hierarchical organization of agents to combine partial information. The hierarchical team organization supports processes for very quickly combining partial results to form an accurate global solution. Each level narrows the uncertainty about the solution based on the data obtained from lower levels. We proved that the hierarchical processing of information reduces the time needed to form the accurate global solution. We tested the performance of the DDM through extensive experimentation in a simulated environment involving many sensors. The simulation models a suite of Doppler sensors used to form a global information map of targets moving in a steady velocity as a function of time. A Doppler sensor is a radar which is based on the Doppler effect. Due to its nature, a Doppler sensor may provide information only about an arc that a detected target may be located on as well as the velocity towards that sensor, that is, the radial velocity [3]. Given a single Doppler measurement, one cannot establish the exact location of a target and its exact velocity; therefore, multiple measurements must be combined for each target. This problem was devised as a challenge problem by the DARPA Autonomous Negotiating Teams (ANTS) program to explore realtime distributed resource allocation algorithms. We compared our hierarchical architecture to other architectures and showed that the monitoring task is faster and more accurate in DDM. We have also shown that DDM can achieve these results when using a low volume of noisy communication.