Thesis

Gameplay in real-time strategy games (RTS) seems to be somehow confined to a de facto standard where economical micro-management is equally important - if not more - as combat strategy. Not all games follow this tendency and some become successful exactly for offering the opposite: a strategy-focused gameplay. However, this is usually achieved by completely removing key gameplay aspects of the genre, such as resource gathering or infrastructures and units? creation, allowing the player to focus on combat. A possible way to create strategy-focused RTS games without sacrificing the other key aspects of the genre is to automate the repetitive and time-consuming tasks. The player will still be in control of everything but he?ll spend less time executing the micro-management tasks. In this work we present a first step towards this solution: an automated system for unit production scheduling that we believe will allow the exploration of new paradigms of play. To be accepted by the player, such system must, among other things, be efficient and reliable, which is a non-trivial task when considering the highly dynamic nature of the environment in this genre of games. To overcome such challenge, we propose a system inspired in the swarm intelligence demonstrated by social insects, namely wasps. The performed evaluation revealed that the proposed decentralized solution yields satisfactory results in comparison with a global centralized algorithm. Although it possesses limitations in particular situations we?ve identified the scenarios in which it performs best, therefore recommending its use if certain problem characteristics are observed.

With the increasing of computers in the daily life of everyone highlights the relationships we develop with them and the importance of these relationships. In order to provide synthetic characters with relationships, it is important to use the psycho-sociological background of the human relationships and strategies used by humans to create and manage those relationships. We have developed a model of relationships that uses the processes involved in these relationships in several dimensions. For this purpose we have based our research in both computational systems that include relationships and in theoretical theories of the psycho-sociology of human relationships. In this way, our model was constructed according to the three main concepts identified in our research: stages: fixed set of levels that different relationships go through; filters: sequence of decisions where we make a choice to maintain or exclude someone from the relationship; strategies: the evolution of a relationship is not a static process but a consequence of interactions by someone and the perception and validation of that interaction by the other relationship partner. As a way to validate and test our model, we have introduced the intimacy relationship and conducted an experiment with eight users where they interacted with a prototype of a game where they to develop a relationship with one character. The evaluation of the experiment results showed that, from all the test conditions, the one that included our model presented the better results, for all the three variables of measure, in the relationship establishment.

In this document, we present a videogame that adapts its content to the player. Such a game needs to infer the player's type from his behavior, and then select how content is managed and presented to the player based on that type. In this work we focus on the later aspect, assuming we already know the player type. We also propose how such information can be used to enhance the player?s experience. After revising the literature on the subject, we decided to use the Demographic Game Design (DGD) model as our player model. Therefore, before playing our game, the player has to fill a questionnaire to assess his Myer-Briggs personality type. From this questionnaire, the game classifies the player according to the DGD model. The game is then adjusted according to this player type, which will influence how the information of the game is presented to the player, in three main aspects: presentation, difficulty management and depth of control over aspects of the game. To evaluate our approach, we asked different types of players to play our game under different conditions and evaluated the experience using a final questionnaire based on the GameFlow model. The evaluation suggests that the player enjoyment is higher when the game is using our framework to adapt to the player.

The objective of this thesis was to create a model for agent memory retrieval of emotionally relevant episodes. Additionally we wished to assess if including such a system in an architecture would improve the agents' believability. We reviewed some core concepts concerning human memory, appraisal theories and believability. Then we analyzed agent architectures that support memory and emotions, realizing that none fulfilled all our requirements for memory retrieval. We proceeded by describing our retrieval model consisting of two main steps: "ecphory", in which the perceived stimuli (retrieval cues) are matched with memories; and "recollective experience" that re-appraises memories that had a positive match. We proposed a location ecphory approximation, in which locations serve as indirect retrieval cues. We described how we implemented our model and used it to drive the behavior of characters in a game application. We implemented the location ecphory approximation, and the recollective experience consisting of a reactive appraisal module with reaction rules for events generated by the former. We recorded the application running and used the videos to create a non-interactive evaluation. The evaluation's structure was a between-groups pre-test/post-test one with two control groups, and we had 96 participants. The evaluation's results are consistent with our hypothesis that agents modeled by our architecture would be perceived as more believable than agents modeled in similar architectures without episodic retrieval.