Agent-Environment Interaction in MAS - Introduction and Survey
Joonas Kesäniemi, Vagan Y. Terziyan · InTech eBooks · 2011
Will-be-set-by-IN-TECH Role of the environment"Agents cannot be considered independently of the environment in which they exist and through which they interact" (Müller, 1996) This chapter elaborates the different aspects of environment in the agent related research literature.According to the survey by Weyns et al (Weyns et al., 2005) one of the main causes of confusion what is actually included to the environment, is brought about by the fact that the environment, as an abstract, logical concept of agent related theories and models, is mixed up with the environment as infrastructure for engineering agent based systems.This of course stems partially from the inherited generality of the term "environment" and the confusion can be alleviated by proper definitions for every given context.However, since one of the goals of this chapter is to contribute to the general discussion of the role of environment in multi-agent systems, we are going to have to mix it up and cover both theoretical and practical sides of the term.The main distinction is made between the artificial intelligence (AI) and agent-oriented software engineering (AOSE) related uses of the environment in an effort to facilitate the discussion about the relationship between agent and its environment in both MAS theory and systems engineering. Agent and its environmentThe "classic" agent research that can be classified under AI discipline has been quite agent centric.The research has concentrated on areas such as agent architectures, agent communication languages and agent coordination models based on direct agent-to-agent communication.In the process, the role of the environment has often remained ambiguous or implicit.One of the classic definitions of agent-hood by Russell and Norvig (Russell & Norvig, 1995) defines agent as being "anything that can be viewed as perceiving its environment through sensors and acting upon that environment through effectors.",which in other words proposes that environment is basically anything outside the boundaries of an individual agent and that all the interaction in multi-agent system happens between agent and its environment.(Russell & Norvig, 1995) also describes some characteristics of the agent environments:• Accessible vs. inaccessible: indicates whether agent can sense the complete state of the world or not.• Deterministic vs. nondeterministic: indicates whether the state changes of the environment are completely determined by its current state or not.• Episodic vs. non-episodic: indicates whether the agent's interaction sequences with its environment (or "episodes") can be though of as independent or not.• Static vs. Dynamic: indicates whether the state of the environment can change while the agent deliberates or not.• Discrete vs. continuous: indicates whether the agent's interactions with the environment are limited or not.Back in the nineteen nineties, environment was generally handled in a application specific manner as demonstrated by examples of agent environment in (Russell & Norvig, 1995), which uses chessboard as an example of simple accessible, deterministic, episodic and discrete environment.Chess pawns are modeled as agents.