It is well known how challenging is the task of coding complex agents for virtual environments. This difficulty in developing and maintaining complex agents has been plaguing commercial applications of advanced agent technology in virtual environments. In this paper, we discuss development of a commercial-grade integrated development environment (IDE) and agent architecture for simulation and training in a high-fidelity virtual environment. Specifically, we focus on two key areas of contribution. First, we discuss the addition of an explicit recipe mechanism to Soar, allowing reflection. Second, we discuss the development and usage of an IDE for building agents using our architecture; the approach we take is to tightly-couple the IDE to the architecture. The result is a complete development and deployment environment for agents situated in a complex dynamic virtual world.
Ari Yakir, Gal A. Kaminka