Christian J Darken

Abstract:  Most first-person shooter game AI's are poor at quickly getting out of lines of fire. AI agents that pass up obvious opportunities to hide or take cover can ruin a game's immersiveness. We will present the sensor grid algorithm that can be used to address this problem, and has been implemented on top of America's Army. The algorithm performs a focused run-time search in the immediate vicinity of the agent. This allows it to be both fast and to react to changes in the environment. Background Taking cover is a universal human response to threat. However, it is not innate; children must learn to hide. It is also not totally understood; psychologists are still investigating a critical part of hiding, which is what we know of what other people can or cannot see (Kelly et. al.). Nonetheless, nearly everyone is able to quickly and effectively duck to safety when threatened. The use of cover is also not purely defensive in nature. A person can be taught to take advantage of cover when moving to make invisible shifts in their position and to minimize their exposure to danger when shooting. The ability to use cover effectively is one of the skills that separate the best real players in first-person shooters from the average players. Unfortunately in the current state of gaming it is also one of the ways to distinguish between live players and game agents. Game agents do not use cover effectively. Typical problems include running right by good sources of cover, and failing to consistently take the most direct route to safety. This paper describes an algorithm that allows software agents to perceive and exploit opportunities to duck out of the sight of a set of ?observers ? (generally, the hostile forces that might shoot at the agent). The goal is realistic short-term ?react to fire ? behavior, rather than constructing a concealed path to a specific goal. Strictly speaking, the algorithm as described finds concealment rather than cover, but extension to