dTank: An Environment for Architectural Comparisons of Competitive Agents

We introduce dTank, a competitive environment, as useful for architectural comparisons of competitive agents and comparisons of human and agent behavior. dTank, a Java-based simulation, was designed to facilitate these forms of holistic comparison of emergent behavior. We present several models built using several cognitive and agent architectures (Java, Jess, and Soar), and compare them against a novice participant. We find that our novice participant used two strategies while playing: wandering and attacking. All three agents accurately modeled the novice’s wandering behavior, while the Soar agent more accurately modeled the novice’s attack behavior. These results demonstrate the usefulness of dTank as a teaching and modeling tool, and as a lightweight environment for testing cognitive models and architectures.