The design of intelligent agents. A layered approach (Q2564761)

This book describes INTERRAP, a pragmatic control architecture for dynamically interacting agents. Based on a layered architecture for integrating reactive and deliberative capabilities of autonomous agents, the notion of procedures is introduced that allow the agent to perform routine tasks efficiently, as well as a cooperative planning layer that is located on top of the local planning layer and that enables an agent to reason about other agents and to coordinate its activites with other agents by explicitly communicating information, goals, and plans. Using such an architecture allows the designer of a multi-agent system to deal with a broad spectrum of interactions such as conflict among agents and collaboration to achieve a common goal by using both local behavior-based mechanisms and mechanisms based on explicit coordination by communication. The book considers the modeling of agents in a multi-agent system from the perspective of the designer of an individual agent. In this agent-centered view, an agent system is looked upon as consisting of an agent and its environment, the latter being described in the internal representation accessible to the agent in a state-based manner. The agent model is illustrated by the simulation of an automated loading dock with autonomous forklift robots. For evaluation purposes, a probabilistic model is presented that allows the agent designer to predict the global effects of a specific class of behavior-based decision algorithms. It is complemented by results of a series of experiments that were carried out using the FORKS simulation system which helps to compare the behavior of agents using behavior-based algorithms for conflict resolution with those relying on coordination through communication.