Multi-Agent Artificial Intelligence in Pursuit Strategies: Breaking through the Stalemate

摘要

The typical artificial intelligence in gaming is single-agent. It is tasked with attacking the playing character and focuses so tightly on this objective that it acts as if it is the only enemy in the game. Typically it does not differentiate being in a setting where it is the only enemy attacking the player and a similar setting where there are multiple agents attacking the player [8]. These multiple agents are acting as single agents and losing their potentially multiplicative effect. This leads to nonsensical and simplistic "tricks" that defeat the game's artificial intelligence (AI) as well as defeating the individual AI agents without having to overcome their strategy. We wish to show that in gaming AI coordinated enemies can significantly improve the gaming experience while maintaining the game designer's original strategic intent. This coordinated multi-agent AI will be shown to have a significant impact on length of play even in simplistic games. While there are several existing methods for multi-agent AI, we present a novel approach that shares information from each individual AI agent with the other agents on their team. This differentiates it from flocking (where the other agents are often treated as additional obstacles to be avoided) and teaming (where the agents focus on the same objective but without the coordinated formational attacks). It is our hypothesis that such information sharing at the individual AI agent level creates a coordinated AI for the overall game that increases the difficulty and challenge of gameplay and requires a better strategy from the player to overcome. In gaming, where a major concern is aesthetics (i.e., how a game makes a user feel), this loss of strategic influence can be devastating to the overall enjoyment of the game. As an example, in the early first person shooter Doom from ID Software[10] the player explored room after cavernous room of a dungeon, each filled with a variety of monsters. The variety of the monsters within these rooms was specifically designed to create an ever-increasing challenge to the player as they entered and had to run around the room while avoiding the enemy, gathering goods, and dispatching the monsters. However, the AI was designed so that each monster independently pursued the player once they entered the room. This resulted in an unintentional "cheat" whereby a player could enter a room, wait one second for each monster to recognize them, and then retreat from the room. This resulted in each of the monsters funneling through a choke point (the door) and made elimination of the threat trivially easy. This result short-circuited the otherwise well-designed gameplay intended by the designers. It is noteworthy that this behavior was lessened in subsequent releases of the game [10]. We wish to present an alternative form AI that avoids this limited type of interaction, namely the AI agents acting independently of each other rather than working together as a team. To do so, we add the multi-agent functionality to the AI for a simple pursuit game. Initially the AI directs each agent independently to pursue the target player. These agents then suffer from collision and overlapping such that the player can evade the clustered agents without difficulty. Next we introduce our multi-agent AI that coordinates the efforts of the enemy agents so that they stay in formation and work together to corner the player. In so doing we wish to show that this greatly improves the quality of gameplay and the realism simulated by the AI. Further, this upholds the original intention of the AI as designed by the developers and avoids unrealistic "cheats" to circumvent the intended gameplay. While this research is centered in gaming, we also believe that it has further reaching applications in security, simulations, and robotics.