Combining Expert Knowledge and Learning from Demonstration in Real-Time Strategy Games

A Survey of Behavior Trees in Robotics and AI

“…BTs for Game AI Type of Game Papers Real time strategy games (RTS) [38], [136], [167], [94], [65], [119], [111], [168], [120], [93] First Person Shooters (FPS) [46], [75], [71], [47], [144], [135], [17] Platform games [7], [177], [109], [129], [37], [178], [32], Dialogue games [102], [36], [133], [42], [158], [79], [34] robot. The parallel node was also discussed and expanded by [35] in combination with event driven nodes, in the context of an architecture for industrial robot control based on BTs.…”

“…Real-time strategy games require players to make tactical decisions, such as taking control of some given area, or aim an attack on some particular enemy facilities. In such games, there is a large number of different units, that can be assigned to specific tasks based on their different strengths and weaknesses [120]. These units can also be combined into squads, platoons and so on up to an entire army, and BTs have been used to make decisions on many such levels of abstraction [38].…”

“…Following this process, rules can be constructed that may solve future problems by retrieving and reusing stored behaviors [46,119]. One drawback of this method has been the difficulty of refining existing strategies that have been previously learned from experience [120], but it has been shown that some additional adaptability can be provided using RL [180].…”

A Review of Real‐Time Strategy Game AI

“…Behavior trees are hierarchies of decision and action nodes that are commonly used by programmers and designers in the game industry in order to define behaviors (effectively a partial plan) for agents (Palma et al 2011). They have become popular because, unlike scripts, they can be created and edited using visual tools, making them much more accessible and understandable to nonprogrammers (Palma et al 2011). Additionally, their hierarchical structure encourages reuse, as a tree defining a specific behavior can be attached to another tree in multiple positions or can be customized incrementally by adding nodes (Palma et al 2011).…”

“…In the case of a CBP system, the solutions found are a set of potential plans or sub‐plans that are likely to be effective in the current situation. CBP systems can exhibit poor reactivity at the strategic level and excessive reactivity at the action level, not reacting to high‐level changes in situation until a low‐level action fails, or discarding an entire plan because a single action failed (Palma et al 2011). One of the first applications of CBP to RTS games was by Aha, Molineaux, and Ponsen (2005), who created a system that extended the dynamic scripting concept of Ponsen et al (2005) to select tactics and strategy based on the current situation.…”