580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can General Video Game AI systems be designed to play a wide variety of games without domain-specific knowledge?
This research area focuses on developing AI agents capable of playing multiple, previously unseen video games without relying on game-specific heuristics or prior training on the game rules. It aims to approximate aspects of general artificial intelligence by creating agents that can adapt to different game dynamics and constraints within strict time limits, often using declarative game description languages and standardized interfaces.
Key finding: The General Video Game AI (GVGAI) framework introduces a Video Game Description Language (VGDL) to define a large space of arcade and puzzle games, enabling AI agents to receive object-oriented state information and select... Read more
Key finding: Expands general game playing from turn-taking board games to real-time video games by defining games declaratively and requiring agents to infer game dynamics and optimal strategies on-the-fly. The paper underlines the need... Read more
Key finding: Provides foundational concepts and algorithms for real-time 3D game AI, including movement algorithms, steering behaviors, collision avoidance, and finite state machines, which underpin generalizable AI agent behaviors. The... Read more
2. What methodologies enable AI-assisted game development, including playtesting, content generation, and level design?
This theme explores AI frameworks and toolkits designed to assist game developers in creating, testing, and refining games through automated agents, procedural content generation, and multi-agent systems. It focuses on how AI can reduce development time, improve game balance, and offer insightful statistical analyses to guide iterative design, thereby bridging academic research and industry practices.
Key finding: Demonstrates that automated AI agents simulating thousands of gameplay sessions can reveal game imbalances, ineffective rewards, and strategic options more quickly than human testers, thereby facilitating efficient... Read more
Key finding: Presents an evaluative study of a multi-agent human-in-the-loop procedural content generation system, showing how computational agents recommending game maps can positively influence human designer decisions, facilitating... Read more
Key finding: Introduces Pogamut 3, an open-source platform that provides out-of-the-box functionality such as sensory-motor primitives, debugging tools, and agent behavior architectures integrated with the Unreal Tournament 2004... Read more
Key finding: Proposes a learning and planning framework tailored to the unique challenges of modern game development, emphasizing AI agent roles beyond winning—such as mimicking human-like behavior and style, assisting developers in... Read more
3. How can AI and machine learning techniques be applied in serious games to model player behavior, personalize experiences, and generate believable non-player characters?
This theme addresses the integration of advanced AI components such as player modeling, emotion recognition, natural language processing, and behavior synthesis into serious and educational games. It emphasizes reusability of AI modules across platforms to enable personalized learning, realistic NPC interactions, and adaptive difficulty, enhancing engagement and educational efficacy.
Key finding: Details the RAGE project that developed a suite of open-source, reusable AI components for serious games, including real-time facial emotion recognition, automated difficulty adaptation, sentiment analysis, and believable NPC... Read more
Key finding: Surveys core AI techniques for serious games focusing on user identification through learning style models and engagement detection, and content adaptation strategies that tailor pedagogical content and non-player character... Read more
Key finding: Examines computational models for NPC behavior grounded in psychology and sociology, categorizing NPCs as scripted, reactive, deliberative, or hybrid; advocates for integrating believable social behaviors and emotional models... Read more
Key finding: Introduces a novel player-centered evaluation approach based on analyzing player-created narrative retellings of gameplay, revealing how AI-driven game mechanics facilitate or hinder meaningful story construction and... Read more