Potential-field based navigation in StarCraft

Abstract

Abstract-Real-Time Strategy (RTS) games are a sub-genre of strategy games typically taking place in a war setting. RTS games provide a rich challenge for both human-and computer players (bots). Each player has a number of workers for gathering resources to be able to construct new buildings, train additional workers, build combat units and do research to unlock more powerful units or abilities. The goal is to create a strong army and destroy the bases of the opponent(s). Armies usually consists of a large number of units which must be able to navigate around the game world. The highly dynamic and real-time aspects of RTS games make pathfinding a challenging task for bots. Typically it is handled using pathfinding algorithms such as A*, which without adaptions does not cope very well with dynamic worlds. In this paper we show how a bot for StarCraft uses a combination of A* and potential fields to better handle the dynamic aspects of the game.

Robustness of coevolved strategies in a real-time strategy game

“…In addition to coevolutionary methods, other online and offline methods have been used to find good RTS strategies, such as case-based reasoning, genetic algorithms, dynamic scripting, and reinforcement learning [10], [11], [12], [13], [14]. Research not only focus' on producing a player to manage an entire RTS game, but also on producing strategies for specific elements of play, such as combat, positioning, navigation, cooperation and resource management [15], [16], [17], [18].…”

Finding robust strategies to defeat specific opponents using case-injected coevolution

“…While some research focuses on producing a player to manage an entire RTS game, other research focuses on finding good strategies for specific RTS game elements, such as combat, positioning, navigation, cooperation and resource management [18], [19], [20], [21].…”

A Survey of Real-Time Strategy Game AI Research and Competition in StarCraft

“…In order to achieve human-level performance, RTS AI designers use a lot of domain knowledge in order to divide the task of playing the game into a collection of sub-problems, which can be dealt-with using individual AI techniques. Figure 3 shows some representative examples of the architectures used by different bots in the AIIDE and CIG StarCraft AI competitions (see Section V): BroodwarBotQ [50], Nova [41], UAlbertaBot [12], Skynet, SPAR, AIUR, and BTHAI [43]. Each box represents an individual module with a clearly defined task (only modules with a black background can send actions directly to StarCraft).…”

“…Some uses of potential fields in RTS games are: avoiding obstacles (navigation), avoiding opponent fire [41], or staying at maximum shooting distance [42]. Potential fields have also been combined with A* path-finding to avoid local traps [43]. Hagelbäck and Johansson [44] presented a multi-agent potential fields based bot able to deal with fog-of-war in the Tankbattle game.…”