Kiting in RTS Games Using Influence Maps

Influence Maps have been successfully used in controlling the navigation of multiple units. In this paper, we apply the idea to the problem of simulating a kiting behavior (also known as “attack and flee”) in the context of real-time strategy (RTS) games. We present our approach and evaluate it in the popular RTS game StarCraft, where we analyze the benefits that our approach brings to a StarCraft playing bot. Introduction Real-time Strategy (RTS) is a game genre where players need to build an economy (gathering resources and building a base) and a military power (training units and researching technologies) in order to defeat their opponents (destroying their army and base). RTS games are real-time, non deterministic, partially observable and have huge state spaces. Therefore, compared to traditional board games, RTS games pose significant challenges for artificial intelligence (Buro 2003). One of such challenges is unit and group control. How to effectively control squads of units or how to simulate complex tactical behaviors in a real time environment is still an open problem in game AI research. Specifically, in this paper, we present an approach based on influence maps to simulate kiting behavior (also known as “attack and flee”). This advanced tactical move is specially helpful to handle combats where we are weaker than our enemy but our attack range is bigger than the enemy. In those cases using a kiting behavior is the difference between losing or wining. There has been a significant amount of work on recreating realistic squad movements. From using the flocking behaviors described by Reynolds (1999) or improving pathfinidng using Influence Maps (Tozour 2001), to combining both techniques (Preuss et al. 2010) for RTS games. Those approaches focus on building a robust navigation system, which is still an open problem for real-time games. To solve this problem several solutions have been proposed in several areas. For example, in the area of robotics, Khatib introduced the concept of Artificial Potential Fields (Khatib 1985) to avoid obstacles in real-time for a robot control. Potential fields and Influence Maps have also been found useful for navigation purposes in the domains of robot soccer Copyright c © 2012, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. (Johansson and Saffiotti 2002), or in computer games such as Quake II (Thurau, Bauckhage, and Sagerer 2000) or RTS games (Hagelbäck and Johansson 2008; Avery, Louis, and Avery 2009). However, all of these approaches give more importance to navigation rather than to tactical offensive or defensive moves. Only in their work on potential fields, Hagelbäck and Johansson showed a simple version of the “attack and flee” behavior, but it was not studied in depth. The work presented in this paper represents a contribution towards achieving more complex and effective tactical moves in real-time domains. Specifically, we propose the use of influence maps (a sister technique to potential fields) to achieve kiting. The approach presented in this paper has been evaluated in the context of StarCraft, and incorporated into the NOVA StarCraft bot (Uriarte 2011) for testing purposes. One of the main disadvantages of potential fields and influence maps is the need to perform parameter tuning. One approach to deal with parameter tuning is to use automatic techniques such as reinforcement learning to converge to good settings (Liu and Li 2008). In this paper we will provide analytic ways to define such parameters whenever possible. The rest of this paper is organized as follows. First we define the problem statement and some background concepts. Then, we briefly describe StarCraft (our domain) and NOVA (the bot into which we have incorporated our approach). After that, we describe our approach to simulate kiting behavior. Then we present an empirical evaluation of the performance of our approach. Finally, the paper closes with conclusions and directions of future research.