A Real-Time Agent Architecture: Design, Implementation and Evaluation

Abstract

The task at hand is the design and implementation of real-time agents that are situated in a changeful, unpredictable, and time-constrained environment. Based on Neisser’s human cognition model, we propose an architecture for real-time agents. This architecture consists of three components, namely perception, cognition, and action, which can be realized as a set of concurrent administrator and worker processes. These processes communicate and synchronize with one another for real-time performance. The design and implementation of our architecture are highly modular and encapsulative, enabling users to plug in different components for different agent behavior. In order to verify the feasibility of our proposal, we construct a multi-agent version of a classical real-time arcade game “Space Invader” using our architecture. In addition, we also test the competitive ratio, a measure of goodness of on-line scheduling algorithms, of our implementation against results from idealized and simplified analysis. Results confirm that our task scheduling algorithm is both efficient and of good solution quality.