Non-player character (NPC) dialogue plays a crucial role in shaping the player experience in narrativedriven video games, influencing agency, immersion and story engagement. Despite the recent advancements in large language models (LLMs) for dynamic dialogue generation, few empirical studies have compared their impact across different dialogue system designs. This pilot study explores how LLM-driven dialogue systems affect the player experience using a custom-developed role-playing game (RPG) featuring four different dialogue designs; static control (CV), rephrase (A), hybrid (B) and fully open-ended (C). Behavioral data and post-game questionnaires were collected from 64 participants. Results indicate that fully open-ended dialogues led to significantly longer dialogue interactions and higher overall engagement, particularly among casual players, with the survey feedback highlighting its immersive and natural tone. These findings suggest that fully open-ended LLM-based dialogue in video games can enhance narrative depth and player involvement.

This study investigates player perceptions of conceptual Large Language Model (LLM)-driven Non-Player Characters (NPCs) in World of Warcraft (WoW), referred to as “AIBuddies.” Building on prior work on Conversational Artificial Autonomous Agents (CA-bots), it explores how such companions could enhance gameplay and player experience. A mixed-method survey of 273 WoW players was conducted, using visual mock-ups to present the AI-Buddy concept rather than a working prototype. Participants evaluated interaction features, customization options, and potential gameplay impact. Results show general support, particularly among Casual players, for using AI-Buddies to improve solo content, personalization, and immersion. However, concerns were raised about balance, multiplayer exploitation, and ethical implications. Thematic analysis also revealed worries about reduced social interaction and technical feasibility. This short study contributes to ongoing discussions on CA-bots and LLM-driven NPCs in MMORPGs and offers early insights into how such features are perceived by the WoW playerbase.

Quest of Aivengarde is a custom-built roleplaying game (RPG) that features a traditional dialogue tree alongside three alternative systems, each incorporating large language models (LLMs) to varying degrees — ranging from rephrasing to fully open-ended conversations. The systems are embedded in a shared game world with consistent narrative and challenges, allowing direct comparison of their design trade-offs. The prototype is intended as a modular testbed for future research, offering a flexible framework to experiment with dialogue models in controlled, interactive settings.

This paper explores the use of Long-Short Term Memory Neural Networks (LSTM-NNs) for analyzing and predicting player trajectories. Utilizing the ESTA dataset, which includes data from professional Counter-Strike matches, the study aims to uncover patterns in player movements and evaluate the accuracy of LSTM-NNs in predicting trajectories. Through systematic experimentation, the model achieved a test Mean Absolute Error (MAE) of 14.44 and Mean Squared Error (MSE) of 477.24, demonstrating strong generalization and alignment between predicted and actual trajectories. The paper outlines a straightforward methodology and experimental setup that can be readily applied to other game genres where player trajectory analysis can offer valuable insights. These findings have significant implications for enhancing player performance, developing advanced game strategies, and improving the overall gaming experience.

Video game designers often view confusion as undesirable, yet it is inevitable, as new players must adapt to new interfaces and mechanics in an increasingly varied and innovative game market, which is more popular than ever. Research suggests that confusion can contribute to a positive experience, potentially motivating players to learn. The state of confusion in video games should be further investigated to gain more insight into the learning experience of play and how it affects the player experience. In this article, we design a study to collect learning-related affects for users playing a game prototype that intentionally confuses the player. We assess the gathered affects against a complex learning model, affirming that, in specific instances, the player experience aligns with the learning experiences. Moreover, we identify correlations between these affects and the Player Experience Inventory constructs, particularly concerning flow experiences.

This paper delves into the adaptability of Proximal Policy Optimization (PPO)-trained agents within dynamic environments. Typically, an agent is trained within a specific environment, learning to maximise reward acquisition and to navigate it effectively. However, alterations to this environment can lead to performance deficiencies. Existing research does not fully elucidate how the training of agents influences their adaptability in different environments and which parameters significantly impact this. This study aims to fill this gap, contributing to the creation of more versatile intelligent agents. The objective of this study is to explore how training agents in various environments affects their adaptability when introduced to unfamiliar environments. To this end, 36 models were trained using 36 different configurations to play a one-versus-one (1v1) Snake game. These models were subsequently compared against each configuration to measure their adaptability. The results reveal that map size substantially affect the adaptability of agents in different environments. Interestingly, the results showed that the most adaptive agents were not those trained on the most expansive and complex environment, but rather the simplest.

This paper introduces a method for developing diverse Non-Player Character (NPC) behaviour through a multiagent genetic algorithm based on Map-Elites. We examine the outcomes of implementing our system in a test environment, with a particular emphasis on the diversity of the evolved agents in the feature space. This research is motivated by how diverse NPCs are an important factor for improving player experience. We show how our multi agent map-elite algorithm is capable of isolating the evolved NPCs in the chosen feature space. Results showed that variation in agent fitness could be predicted with $40 \%$ from agent genomes, when agents played 100 games each.

This project presents an evolutionary algorithm for texture generation that allows users to choose and manipulate camouflage patterns. The initial results of a pilot study provide some insight into usability and the users’ ability to replicate a target pattern. The result is an evaluation of gathered data showing user tendencies and how they engage with the system. These tendencies include significantly different completion times for target patterns varying in complexity. Additionally, participants mostly agreed that the tool is helpful for future games and objects other than camouflage skins. The findings suggest potential applications for artificial intelligence in enhancing user customization and design flexibility. Further research must address technical limitations and explore broader game industry implications.

Camouflage serves a dual purpose in games, imparting believability of the world and narrative and being a game asset used by players for customization. Using the U.S. Woodland Battle Dress Uniform or M81 as an inspiration for the pattern, we present a system for generating camouflage using a process based on present military evaluations. The proposed approach is a genetic algorithm that uses image processing/analysis techniques to assess the generated texture and can create camouflage that allows the soldier model to blend with the environment. A human evaluation was conducted and shows that the fitness metric used by the system is a suitable surrogate.

This paper describes and illustrates a system designed as part of a submission to the Generative Design in Minecraft (GDMC) competition. It introduces an approach to partitioning of a three-dimensional game space novel to the domain of Minecraft settlement generation: traversal segmentation. Moreover, the paper introduces a novel implementation of a two-system brain model for autonomous agent simulation. Traversal segmentation is used in conjunction with a grid-wise segmentation method to produce a contextual representation of the game space. This is used as input for the settlement generation using autonomous agents, where each agent is controlled by their system 1 impulses, and their system 2 reasoning-action brain model. The two-system brain model is novel to autonomous agent simulation and is described both theoretically and by its implementation. The resulting settlements, generated by settlers upon grid-wise segmentation of the traversable space, boasts the properties; organic settlement evolution, and adaptability to its surrounding terrain, though not realism when compared to settlements procedurally generated by Minecraft.

Boardgames exist that explicitly address Computational Thinking (CT for short) concepts and practices. Some are actual games, while others are more akin to gamified learning activities. And since CT has been formalized only recently, many existing boardgames unknowingly might support aspects of CT. To help educators and game practitioners navigate this complex landscape, we analyze a selected sample of analog games, and propose to categorize their features with respect to CT concepts and practices. The main contribution of this paper is a novel way to identify potential CT-relevant games, that leverages on the authors’ experience with digital and analog games, playful and game-based learning. Although limited, this approach appears promising and practical for CT teachers and game designers interested in adapting existing games to the classroom or developing better CT-supporting boardgames.

This paper describes an evaluation conducted on the MetaCompose music generator, which is based on evolutionary computation and uses a hybrid evolutionary technique that combines FI-2POP and multi-objective optimization. The main objective of MetaCompose is to create music in real-time that can express different mood-states. The experiment presented here aims to evaluate: (i) if the perceived mood experienced by the participants of a music score matches intended mood the system is trying to express and (ii) if participants can identify transitions in the mood expression that occur mid-piece. Music clips including transitions and with static affective states were produced by MetaCompose and a quantitative user study was performed. Participants were tasked with annotating the perceived mood and moreover were asked to annotate in real-time changes in valence. The data collected confirms the hypothesis that people can recognize changes in music mood and that MetaCompose can express perceptibly different levels of arousal. In regards to valence we observe that, while it is mainly perceived as expected, changes in arousal seems to also influence perceived valence, suggesting that one or more of the music features MetaCompose associates with arousal has some effect on valence as well.