With the services that autonomous robots are to provide becoming more demanding, the states that the robots have to estimate become more complex. In this article, we develop and analyze a probabilistic, vision-based state estimation method for individual, autonomous robots. This method enables a team of mobile robots to estimate their joint positions in a known environment and track the positions of autonomously moving objects. The state estimators of different robots cooperate to increase the accuracy and reliability of the estimation process. This cooperation between the robots enables them to track temporarily occluded objects and to faster recover their position after they have lost track of it. The method is empirically validated based on experiments with a team of physical robots.

We present methods that generate cooperative strategies for multivehicle control problems using a decomposition approach. By introducing a set of tasks to be completed by the team of vehicles and a task execution method for each vehicle, we decomposed the problem into a combinatorial component and a continuous component. The continuous component of the problem is captured by task execution, and the combinatorial component is captured by task assignment. In this paper, we present a solver for task assignment that generates near-optimal assignments quickly and can be used in real-time applications. To motivate our methods, we apply them to an adversarial game between two teams of vehicles. One team is governed by simple rules and the other by our algorithms. In our study of this game we found phase transitions, showing that the task assignment problem is most difficult to solve when the capabilities of the adversaries are comparable. Finally, we implement our algorithms in a multi-level architecture with a variable replanning rate at each level to provide feedback on a dynamically changing and uncertain environment. 1

Knowledge and the way it is represented have a tremendous impact on the capabilities and performance of intelligent systems. There is evidence from studies of human cognitive functions that experts use multiple representations in problem solving tasks and know when to switch between representations. In this paper, we discuss the issues pertaining to what types of knowledge are required for an intelligent system, how to evaluate the knowledge and representations, and provide examples of how representation affects and even enables functionality of a system. We describe an example of an intelligent system architecture that is built upon multiple knowledge types and representations and has been applied to a variety of real-time intelligent systems.

Computing is beginning to change as programs start to execute over many mobile processors communicating over ad hoc networks. Collections of these processors can be described as a "swarm." The behavior of a swarm is categorized as the total behavior of all its individual components but, unlike traditional distributed programming, swarms exist dynamically in unpredictable environments. The major challenges are designing programs for the units with a desired swarm behavior and, on the other side, predicting behavior from the programs running on the units. The soccer simulation competition within the RoboCup 2001 conference is the medium of the swarm research. This conference uses a soccer simulation to focus on cooperation between autonomous agents in dynamic multiagent environments. The simulation league comprises of a server acting as the field, and eleven clients for each team, which act as the players. The field is an unpredictable dynamic environment, while the players are thought of as the cooperative swarm. The research addresses the challenges of swarms by implementing an extendable object-oriented architecture for a RoboCup soccer player. Testing the ease of adding the centering and dispersing defensive behaviors displays the benefits of the program architecture. The extendable object-oriented design resulted in easily implementing the two behaviors into the swarm program of the RoboCup soccer player. Though RoboCup is only one application of swarm programming, the architecture can be applied to many others. If utilized, the swarm development community could evolve more effectively with endless potential. iii iiiiii iii Table of Contents ABSTRACT ........................................................................................................................

We use a decomposition approach to generate cooperative strategies for a class of multi-vehicle control problems. By introducing a set of tasks to be completed by the team of vehicles and a task execution method for each vehicle, we decompose the problem into a combinatorial component and a continuous component. The continuous component of the problem is captured by task execution, and the combinatorial component is captured by task assignment. In this paper, we present a solver for task assignment that generates near-optimal assignments quickly and can be used in real-time applications. To motivate our methods, we apply them to an adversarial game between two teams of vehicles. One team is governed by simple rules and the other by our algorithms. In our study of this game we found phase transitions, showing that the task assignment problem is most difficult to solve when the capabilities of the adversaries are comparable. Finally, we utilize our algorithms in a hierarchical model predictive control architecture with a variable replanning rate at each level to provide feedback in dynamically changing and uncertain environments. c ○ 2006 Elsevier B.V. All rights reserved.