Approximate dynamic programming with applications in multi-agent systems

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2007.This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.MIT Institute Archives copy: contains CDROM of thesis in .pdf format.Includes bibliographical references (p. 151-161).This thesis presents the development and implementation of approximate dynamic programming methods used to manage multi-agent systems. The purpose of this thesis is to develop an architectural framework and theoretical methods that enable an autonomous mission system to manage real-time multi-agent operations. To meet this goal, we begin by discussing aspects of the real-time multi-agent mission problem. Next, we formulate this problem as a Markov Decision Process (MDP) and present a system architecture designed to improve mission-level functional reliability through system self-awareness and adaptive mission planning. Since most multi-agent mission problems are computationally difficult to solve in real-time, approximation techniques are needed to find policies for these large-scale problems. Thus, we have developed theoretical methods used to find feasible solutions to large-scale optimization problems. More specifically, we investigate methods designed to automatically generate an approximation to the cost-to-go function using basis functions for a given MDP. Next, these these techniques are used by an autonomous mission system to manage multi-agent mission scenarios. Simulation results using these methods are provided for a large-scale mission problem. In addition, this thesis presents the implementation of techniques used to manage autonomous unmanned aerial vehicles (UAVs) performing persistent surveillance operations. We present an indoor multi-vehicle testbed called RAVEN (Real-time indoor Autonomous Vehicle test ENvironment) that was developed to study long-duration missions in a controlled environment.(cont.) The RAVEN's design allows researchers to focus on high-level tasks by autonomously managing the platform's realistic air and ground vehicles during multi-vehicle operations, thus promoting the rapid prototyping of UAV technologies by flight testing new vehicle configurations and algorithms without redesigning vehicle hardware. Finally, using the RAVEN, we present flight test results from autonomous, extended mission tests using the technologies developed in this thesis. Flight results from a 24 hr, fully-autonomous air vehicle flight-recharge test and an autonomous, multi-vehicle extended mission test using small, electric-powered air vehicles are provided.by Mario J. Valenti.Ph.D