Dynamic difficulty adjustment for skill acquisition in games

When used in an educational or training setting, video games have been shown to be an effective knowledge transfer and skill acquisition method. This field of study is known as serious games. In this context, video games aim to tackle some of the shortcomings of traditional classroom teaching, namely student engagement and motivation. However, video games can still fail to motivate and engage students if the level of challenge presented by the video game is not at the right level for the student. Additionally, measuring the learning outcomes can be difficult without further testing or direct competition. In the current literature, there is no unified framework that brings together the concepts of providing the right level of challenge for a player, in conjunction with an appropriate way to measure and monitor their outcomes. In this thesis, we present a generalised training framework to address this gap in the literature. The framework will consist of novel adaptive gameplay agents, and new measuring techniques to improve knowledge transfer and skill acquisition in serious games.  Currently, in the video game industry, the level of challenge presented by most games is through the selection of difficulty level by the player before gameplay. Some games offer no difficulty selection, only providing one difficulty level for all players. Previous research has shown players will stop playing a video game that is too difficult, conversely, they will also stop playing if they become bored because the video game is too easy. If players are challenged at an appropriate level compared to their skill, they can become more engaged and motivated, entering into a state of flow. Dynamic Difficulty Adjustment (DDA) is a field of study that aims to provide the player with a consistent level of challenge that adapts to their current skill level through real-time control of the artificially intelligent opponents or game environment. Understanding the learning outcomes of the player is also an important aspect of the application of serious games. As the learning outcomes can be used to show the effectiveness of the serious game, while also providing the trainers/educators with a way of monitoring the player's progress. Whilst adaptive agents can be used to improve engagement, and subsequently flow and immersion resulting in improved learning outcomes for players, there is currently no unified framework under which Dynamic Difficulty Adjustment agents can be utilised in conjunction with a method to effectively measure the learning outcomes of the player.     This thesis presents the Adaptive Training Framework (ATF) as a framework by which trainers and teachers can effectively implement knowledge transfer and skill acquisition through video games, through the application of novel Dynamic Difficulty Adjustment agents and a novel approach to measuring the skill level of the players. The ATF consists of three phases, with the first Phase being adaptive gameplay. In this phase, we implement the concepts and use of adaptive agents in gameplay. As part of this research, we also present six novel Monte Carlo Tree Search (MCTS) DDA agents, which can be implemented to control artificially intelligent opponents. They can be utilised to provide the player with a constant level of perceived challenge. These agents can also be modified such that the constant level of perceived challenge can be increased or decreased based on the individual player's skill acquisition rate. Due to their ability to provide the player with an engaging and challenging opponent these agents are ideal for use in the ATF during gameplay. Using two different game genres as a testbed, the DDA agents are evaluated through bot and human trials. The experiments show that these novel DDA agents have improved performance in challenge, enjoyment, and realism, when compared to current approaches and can be generalised across game genres. The second Phase is the measurement of the player's skill level. Understanding how the players are achieving their learning outcomes is an important part of serious games. An effective way to measure the players learning objectives will not only allow educators/trainers to know the players are learning, but also measure the effectiveness of the serious game. To be utilised in this phase we present two novel approaches to measuring a player's skill level through only the interaction with the DDA agents. The current state-of-the-art approaches to measuring player skill level require the players to compete against each other to determine their skill level. However, this may not be suitable in a classroom environment and adds a layer of unnecessary and potentially prolonged gameplay. The measuring approaches outlined in this thesis can be utilised as a part of the ATF, to ensure that the player's skill level and skill acquisition rate are accurately measured. The measuring approaches are evaluated against the current state-of-the-art methods through bot trials and confirmed with results from the human trials. The results of these experiments show that through interaction between the player and the DDA agent the effectiveness of learning outcomes can be measured. The final Phase is monitoring the player's skill or knowledge acquisition rate. In this research, we show that different players will learn at different rates when faced with an opponent that has an equal chance of winning or losing the game. The phase in the ATF provides a mechanism by which it can adjust the level of challenge, either by increasing or decreasing the difficulty they present to the player, and terminating training once the desired outcomes are achieved.   This thesis brings together novel advancements in the field of Dynamic Difficulty Adjustment and player skill level measurement within a unified framework. With the goal of improving Serious Games, such that when adaptive games are employed, their effectiveness in both player engagement and learning outcomes are linked and ultimately be utilised to improve player training through serious games.