Comparative Analysis: Top-Down and Bottom-Up Methodologies for Multi-Agent System Design

Topics: Top-down and bottom-up design, Control theory, Software development process Pages: 7 (1931 words) Published: September 2, 2013
Comparative Analysis of Top–Down and Bottom–up Methodologies for Multi–Agent System Design Valentino Crespi
Dept. Computer Science Cal State Los Angeles Los Angeles, CA

Aram Galstyan
Information Sciences Institute Univ. of Southern California Marina del Rey, CA

Kristina Lerman
Information Sciences Institute Univ. of Southern California Marina del Rey, CA

vcrespi@calstatela.edu ABSTRACT

galstyan@isi.edu

lerman@isi.edu

Traditionally, top-down and bottom-up design approaches have competed with each other in Algorithmics and Software Engineering. In the top-down approach, design process starts with specifying the global system state and assuming that each component has global knowledge of the system, as in a centralized approach. The solution is then decentralized by replacing global knowledge with communication. In the bottom-up approach, on the other hand, the design starts with specifying requirements and capabilities of individual components, and the global behavior is said to emerge out of interactions among constituent components and between components and the environment. In this paper we present a comparative study of both approaches with particular emphasis on applications to multi–agent system engineering.

Categories and Subject Descriptors
I.2.11 [Distributed Artificial Intelligence]: Multiagent systems

the properties of a classical centralized solution to the global specification are expected to hold, up to some tolerable performance degradation, also in a decentralized environment. In the bottom-up methodology, on the other hand, the rules of agent interactions are typical. In systems designed starting from the bottom, the global state of all the components is assumed to be impossible to obtain, and the desired collective behavior is said to emerge from interactions among individual agents and between the agents and the environment. In summary, in the top-down design the final distributed solution is obtained as a process of relaxation of the constraints that require instant access to remote resources with infinite precision. The bottom-up design starts with a rigorously pre–decided set of rules for the individual behaviors and local interactions and then proceeds with the inference of the global emergent behavior. While the question of which design is appropriate for a given system extends over the most diverse areas in computer science and computer engineering, in this paper we compare two approaches in a typical domain of multi-agent systems engineering.

General Terms
Design, Algorithms

2. FOUNDATIONS OF THE DESIGN 2.1 Top Down
The top-down methodology has been recently developed to produce provably performant designs relative to what is achieved in classical centralized control theory. This approach has been also successfully applied to problems of sensor localization, distributed vehicle flow control and distributed surveillance [1, 2]. Ideally the designer should start from the definition of an objective that involves global quantities, then devise a centralized optimization algorithm and finally proceed to the synthesis of the decentralized (agent-based) solution. The design process consists of three steps: modeling, synthesis and analysis/optimization. Modeling: In this phase the designer identifies and categorizes system’s agents according to the following taxonomy derived from classical Control Theory. Information agents gather information about the environment and allow its dissemination (analogous to “sensors” in classical systems theory). Modeling agents collect data from many information agents and update internal estimates of the “real world” state (analogous to state estimators, like Kalman filters, in classical systems theory). Planning agents use the current world state estimates, the viable action or control options and the current goals to plan new actions to carry out. These agents may need to task brokering agents (that have no conterpart in classical control theory)...

References: [1] V. Crespi and G. Cybenko. Decentralized Algorithms for Sensor Registration. In Proceedinds of the 2003 International Joint Conference on Neural Networks (IJCNN2003), Portland, Oregon, July 2003. [2] V. Crespi, G. Cybenko, D. Rus, and M. Santini. Decentralized Control for Coordinated flow of Multiagent Systems. In Proceedings of the 2002 World Congress on Computational Intelligence. Honolulu, Hawaii, May 2002. [3] K. Lerman and A. Galstyan. Mathematical model of foraging in a group of robots: Effect of interference. Autonomous Robots, 13(2):127–141, 2002. [4] K. Lerman, A. Galstyan, A. Martinoli, and A. Ijspeert. A macroscopic analytical model of collaboration in distributed robotic systems. Artificial Life Journal, 7(4):375–393, 2001.
1160
Continue Reading

Please join StudyMode to read the full document

You May Also Find These Documents Helpful

  • Essay about Top Down and Bottom Up Approach
  • Top-Down, Bottom-Up in Cognition Essay
  • Comparative analysis of File System of Operating System Essay
  • Top Down vs Bottom Up Change Essay
  • Top-Down Versus Bottom-Up Analysis Essay
  • Top Down Leadership vs Bottom Up Leadership Essay
  • Peception Involves Bottom and Top Down Processing Essay
  • Essay about System Design Methodology

Become a StudyMode Member

Sign Up - It's Free