Dynamic Game Difficulty Balancing - Approaches

Approaches

Different approaches are found in the literature to address dynamic game difficulty balancing. In all cases, it is necessary to measure, implicitly or explicitly, the difficulty the user is facing at a given moment. This measure can be performed by a heuristic function, which some authors call "challenge function". This function maps a given game state into a value that specifies how easy or difficult the game feels to the user at a specific moment. Examples of heuristics used are:

• The rate of successful shots or hits
• The numbers of won and lost pieces
• Life points
• Evolution
• Time to complete some task

... or any metric used to calculate a game score.

Hunicke and Chapman’s approach controls the game environment settings in order to make challenges easier or harder. For example, if the game is too hard, the player gets more weapons, recovers life points faster, or faces fewer opponents. Although this approach may be effective, its application can result in implausible situations. A straightforward approach is to combine such "parameters manipulation" to some mechanisms to modify the behavior of the non-player characters (NPCs) (characters controlled by the computer and usually modeled as intelligent agents). This adjustment, however, should be made with moderation, to avoid the 'rubber band' effect. One example of this effect in a racing game would involve the AI driver's vehicles becoming significantly faster when behind the player's vehicle, and significantly slower while in front, as if the two vehicles were connected by a large rubber band.

A traditional implementation of such an agent’s intelligence is to use behavior rules, defined during game development. A typical rule in a fighting game would state "punch opponent if he is reachable, chase him otherwise". Extending such an approach to include opponent modeling can be made through Spronck et al.′s dynamic scripting, which assigns to each rule a probability of being picked. Rule weights can be dynamically updated throughout the game, accordingly to the opponent skills, leading to adaptation to the specific user. With a simple mechanism, rules can be picked that generate tactics that are neither too strong nor too weak for the current player.

Andrade et al. divides the DGB problem into two dimensions: competence (learn as well as possible) and performance (act just as well as necessary). This dichotomy between competence and performance is well known and studied in linguistics, as proposed by Noam Chomsky. Their approach faces both dimensions with reinforcement learning (RL). Offline training is used to bootstrap the learning process. This can be done by letting the agent play against itself (selflearning), other pre-programmed agents, or human players. Then, online learning is used to continually adapt this initially built-in intelligence to each specific human opponent, in order to discover the most suitable strategy to play against him or her. Concerning performance, their idea is to find an adequate policy for choosing actions that provide a good game balance, i.e., actions that keep both agent and human player at approximately the same performance level. According to the difficulty the player is facing, the agent chooses actions with high or low expected performance. For a given situation, if the game level is too hard, the agent does not choose the optimal action (provided by the RL framework), but chooses progressively less and less suboptimal actions until its performance is as good as the player’s. Similarly, if the game level becomes too easy, it will choose actions whose values are higher, possibly until it reaches the optimal performance.

Demasi and Cruz built intelligent agents employing genetic algorithms techniques to keep alive agents that best fit the user level. Online coevolution is used in order to speed up the learning process. Online coevolution uses pre-defined models (agents with good genetic features) as parents in the genetic operations, so that the evolution is biased by them. These models are constructed by offline training or by hand, when the agent genetic encoding is simple enough.

Other work in the field of DGB is based on the hypothesis that the player-opponent interaction—rather than the audiovisual features, the context or the genre of the game—is the property that contributes the majority of the quality features of entertainment in a computer game. Based on this fundamental assumption, a metric for measuring the real time entertainment value of predator/prey games was introduced, and established as efficient and reliable by validation against human judgment.

Further studies by Yannakakis and Hallam have shown that artificial neural networks (ANN) and fuzzy neural networks can extract a better estimator of player satisfaction than a human-designed one, given appropriate estimators of the challenge and curiosity (intrinsic qualitative factors for engaging gameplay according to Malone) of the game and data on human players' preferences. The approach of constructing user models of the player of a game that can predict the answers to which variants of the game are more or less fun is defined as Entertainment Modeling. The model is usually constructed using machine learning techniques applied to game parameters derived from player-game interaction and/or statistical features of player's physiological signals recorded during play. This basic approach is applicable to a variety of games, both computer and physical.