The World's Yogurts

Behavior Design Patterns: Engineering Human Behavior Models
Glenn Taylor Robert E. Wray, PhD Soar Technology, Inc. 3600 Green Court Suite 600 Ann Arbor, MI 48105 734-327-8000 glenn@soartech.com, wray@soartech.com

Presented at Behavior Representation in Modeling and Simulation (BRIMS) May 2004

Keywords: Intelligent Systems, Human Behavior Models, Computer Generated Forces, Agent-basedsystems, Design Patterns ABSTRACT: This paper describes new research that attempts to address cost and development time bottlenecks in the implementation of high-fidelity human behavior representations. The proposed solution, behavior design patterns (BDPs), is an extension of design patterns in traditional software engineering. A crucial challenge for BDPs, and the fundamental limitation of existingdesign patterns, is capturing the unique computational properties of human behavior models (HBMs), which require parallel, goal-oriented execution to achieve autonomy, flexibility and adaptation in task performance. We demonstrate example patterns extracted from an existing HBM, TacAir-Soar, and describe how they can be applied in other HBMs. Although the results are preliminary, one advantage to theproposed approach is the development of solution catalogs, which can be used to improve development today, even while more powerful development tools are being designed and constructed.

1. Introduction
Developing high-fidelity models of human behavior is a challenging endeavor, often resulting in high costs and long development times. High costs are, in part, the result of little formal reuseof software in building these models. Obviously, reuse is difficult when the modeled tasks are very different (e.g., armored cavalry vs. fixed wing aircraft). In addition, differences underlying implementation languages, cognitive architectures, and simulation environments can also contribute to the lack of reuse. Cognitive architectures attempt to capture some of the regularities in human behaviorat the level of symbolic processing and subsymbolic tuning of behavior to the task [1, 2]. Our goal in this work is to extend reuse to encoding common patterns of human knowledge-level behavior. Because such a goal is obviously open-ended, we focus specifically on military domains, where training and common doctrine enforce and reinforce patterns of human behavior [3]. The inspiration for this workis in traditional software engineering techniques of design patterns that have proven useful in building large software systems. Design patterns capture and generalize successful designs for common problems. We are extending the notion of a design pattern to address the challenges of building human behavior representations. In doing so, we examine some of the 1

qualities of intelligent systemsthat make them unique among software systems, and how these qualities might inform new ways to describe designs for intelligent systems. We present an analysis of an existing human behavior model, TacAir-Soar [4, 5], in an effort to demonstrate the prevalence of design patterns in implemented systems.

2. From Design Patterns to Behavior Design Patterns
This section describes the softwareengineering notion of design patterns and how they have been used in traditional software development. Patterns for human behavior representations (HBRs) must be different from traditional design patterns, because the computational support for traditional object-oriented systems is inadequate for the requirements of autonomous, robust, high fidelity HBRs. We introduce Behavior Design Patterns (BDPs),our proposed adaptation of design patterns to capture the different computational infrastructures needed for HBRs. 2.1 Design Patterns In typical software engineering, design patterns represent concise solutions to common problems experienced in the development of large software systems – essentially, they embody a technique to foster reuse of software solutions [6]. Reuse can take many forms,...