What's the Game, then? Opportunities and Challenges for Runtime Behavior Generation
Open accessThe DOI matches its Crossref record; 20 of the 20 references it cites were found in a registry or an index.
Abstract
Procedural content generation (PCG), the process of algorithmically creating game components instead of manually, has been a common tool of game development for decades. Recent advances in large language models (LLMs) enable the generation of game behaviors based on player input at runtime. Such code generation brings with it the possibility of entirely new gameplay interactions that may be difficult to integrate with typical game development workflows. We explore these implications through GROMIT, a novel LLM-based runtime behavior generation system for Unity. When triggered by a player action, GROMIT generates a relevant behavior which is compiled without developer intervention and incorporated into the game. We create three demonstration scenarios with GROMIT to investigate how such a technology might be used in game development. In a system evaluation we find that our implementation is able to produce behaviors that result in significant downstream impacts to gameplay. We then conduct an interview study with n=13 game developers using GROMIT as a probe to elicit their current opinion on runtime behavior generation tools, and enumerate the specific themes curtailing the wider use of such tools. We find that the main themes of concern are quality considerations, community expectations, and fit with developer workflows, and that several of the subthemes are unique to runtime behavior generation specifically. We outline a future work agenda to address these concerns, including the need for additional guardrail systems for behavior generation.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Citations
Cites
Understanding natural language
Winograd, T. (1972) · Cognitive Psychology · doi:10.1016/0010-0285(72)90002-3 · Cited by 2,799 · cited 1×
“Attempts to achieve scene understanding and manipulation date back to 1968 with the SHRDLU system [33], where its users could move colored blocks using natural language.”
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology · doi:10.1145/3586183.3606763 · Cited by 2,187 · cited 1×
“While not specifc to games, Park et al.’s Generative Agents sys-tem also tackles the problem of piloting characters through a virtual environment called Smallville, and takes the approach of using LLMs to directly manage Smallville as well as the agents [23].”
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J. C., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · arXiv · doi:10.48550/arxiv.2304.03442 · Cited by 2,173 · cited 1×
“While not specifc to games, Park et al.’s Generative Agents sys-tem also tackles the problem of piloting characters through a virtual environment called Smallville, and takes the approach of using LLMs to directly manage Smallville as well as the agents [23].”
Procedural content generation for games: A survey
Hendrikx, M., Meijer, S., Van Der Velden, J., & Iosup, A. (2013) · ACM Transactions on Multimedia Computing, Communications, and Applications · doi:10.1145/2422956.2422957 · Cited by 558 · cited 1×
“Games such as Minecraft [21] and No Man’s Sky [14] use procedural generation to create entire virtual worlds for players to discover.”
Procedural Content Generation in Games
Shaker, N., Togelius, J., & Nelson, M. J. (2016) · Computational Synthesis and Creative Systems · doi:10.1007/978-3-319-42716-4 · Cited by 519 · cited 2×
“When applied to behaviors, these approaches are necessarily restrictive in scope due to their inability to generate arbitrary computer programs [26].”
An experiment in automatic game design
Togelius, J., & Schmidhuber, J. (2008) · 2008 IEEE Symposium On Computational Intelligence and Games · doi:10.1109/cig.2008.5035629 · Cited by 246 · cited 2×
“Togelius and Schmidhuber used a discrete 15 by 15 grid populated with a player controlled agent and various colored objects, then used an evolutionary system to create games based on the grid layout and object behavior [29].”
Learning 3D Semantic Scene Graphs From 3D Indoor Reconstructions
Wald, J., Dhamo, H., Navab, N., & Tombari, F. (2020) · 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr42600.2020.00402 · Cited by 244
LLMR: Real-time Prompting of Interactive Worlds using Large Language Models
Torre, F. D. L., Fang, C. M., Huang, H., Banburski, A., Amores, J., & Lanier, J. (2024) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3613904.3642579 · Cited by 232
Browne, C., & Maire, F. (2010) · IEEE Transactions on Computational Intelligence and AI in Games · doi:10.1109/tciaig.2010.2041928 · Cited by 219
Graph-to-3D: End-to-End Generation and Manipulation of 3D Scenes Using Scene Graphs
Dhamo, H., Manhardt, F., Navab, N., & Tombari, F. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.01604 · Cited by 126 · cited 2×
“SSGs provide adaptivity to 3D interaction tasks [10], allow semantic control over generated content [11], and function in a way to generate and manipulate 3D scenes [32].”
Research works printed but not linked: 5
- Beth Elderkin. 2024. GDC 2024 state of the game industry: DEVS discuss layoffs Generative AI and more: News: GDC: Game Developers Conference. https://gdconf.com/news/gdc-2024-state-game-industry-devs-discuss-layoffs-generative-ai-and-more%3F_mc%3Dedit_gdcsf_gdcsf_le_x_17_x_2024%26kcode%3DBLG_GDC
- Huayang Li, Yixuan Su, Deng Cai, Yan Wang, and Lemao Liu. 2022. A survey on retrieval-augmented text generation. arXiv preprint arXiv:2202.01110 (2022). · arXiv
- David McNeill. 1992. Hand and mind: What gestures reveal about thought. University of Chicago press.
- Michael Nitsche, Calvin Ashmore, Will Hankinson, Robert Fitzpatrick, John Kelly, and Kurt Margenau. 2006. Designing procedural game spaces: A case study. Proceedings of FuturePlay 2006 (2006).
- Bernhard Rieder. 2018. Using procedural content generation via machine learning as a game mechanic. Austrian Marshall Plan Foundation (2018).
Cited by
GPT for Games: An Updated Scoping Review (2020–2024)
Yang, D., Kleinman, E., & Harteveld, C. (2025) · IEEE Transactions on Games · doi:10.1109/tg.2025.3563780 · Cited by 53
“In the following sections, we will detail five key applications of GPT in gaming—PCG, MIGDD, MIG, PG, and GUR—highlighting new findings from 2024 and how they differ from earlier works.”
Vachha, C., Kang, Y., Dive, Z., Chidambaram, A., GUPTA, A. C., Jun, E., & Hartmann, B. (2025) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3706598.3714312 · Cited by 25
“These developments have indicated the importance of object-centric labeling and editing in the systems that are built and have guided our design of editing components of 3D scenes in VR.”
Self-citationJust-In-Time Objectives: A General Approach for Specialized AI Interactions
Lam, M. S., Shaikh, O., Xu, H., Guo, A., Yang, D., Heer, J., Landay, J. A., & Bernstein, M. S. (2026) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3790713 · Cited by 30
Empowering Children to Create AI-Enabled Augmented Reality Experiences
Zhang, L., Zhou, S., Liaqat, A., Mak, T. C. W., Berengard, B., Qian, E., & Monroy‐Hernández, A. (2025) · ACM Symposium on User Interface Software and Technology (UIST) · doi:10.1145/3746059.3747662 · Cited by 7
AI for Games in the Foundation Model EraPreprint
Luo, M., Li, Y., Li, H., Lin, H., Zhou, P., Ju, T., Zhang, R., Jin, Y., Lee, M. L., & Hsu, W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.16679 · Cited by 7
HOICraft: In-Situ VLM-based Authoring Tool for Part-Level Hand-Object Interaction Design in VR
Lee, D., Sun, Q., & Yoon, S. H. (2026) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3790823 · Cited by 6
“To this end, prior work has explored automating content creation through text-to-scene generation [2, 22, 24, 48, 74, 84] and text-to-interaction through code synthesis [15, 31, 42, 49].”
Leveling up Learning: Serious Games for Computing Education - Long-Term Opportunities and Risks
Wuensche, B. C., Hooper, S., Whalley, J. L., Straand, I. J., Denny, P. C., Crow, T. V. H., Lange-Nawka, D. C., Luxton-Reilly, A., & Thompson, S. E. R. (2025) · Australasian Computing Education Conference · doi:10.1145/3716640.3716655 · Cited by 5
“…GenAI algorithms, popularised by image generators such as Dall-e, Stable Diffusion, and Mid-Journey, are now able to create convincing looking 2D and 3D assets and video content [7, 46, 68] and recent research has explored the use of LLMs to generate novel game behaviour based on player input [60].”
Democratizing Game Modding with GenAI: A Case Study of StarCharM, a Stardew Valley Character Maker
Miralvand, H. Z., Nikghalb, M. R., Darandeh, M., Khan, A., Arawjo, I., & Cheng, J. (2025) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3748612 · Cited by 5
“Jennings et al. [21] also proposed GROMIT, which is an LLM-based runtime behavior generation system that can be triggered by player action.”
MCP-Unity: Protocol-Driven Framework for Interactive 3D Authoring
Wu, S., & Barnett, J. P. (2025) · SIGGRAPH Asia Technical Communications · doi:10.1145/3757376.3771417 · Cited by 4
“Recent breakthroughs, such as WonderWorld [Yu et al. 2025] et al. 2024] and GROMIT [Jennings et al. 2024] are two pio-neering systems that generate codes at runtime to create and modify interactive experiences in Unity, and DreamGarden [Earle et al. 2025] adapts a similar prompt-to-scene workflow…”
RPGAgent: Driving Coherent Story-to-Play Generation with an LLM-Based Multi-Agent System
Zhang, S., Xiao, Y., Ma, R., & Leung, C.-S. (2026) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3790326 · Cited by 8
“Focusing on gameplay mechanics, Jennings et al. [31] demonstrated that runtime behavior generation can be embedded directly into game mechanics to enable emergent interaction.”
Supporting Students in Prototyping AI-backed Software with Hosted Prompt Template APIs
Aveni, T. J., Smith, J., Fox, A., & Hartmann, B. (2025) · ACM Conference on Innovation and Technology in Computer Science Education (ITiCSE) · doi:10.1145/3724363.3729109 · Cited by 4
Self-citationSceneGenA11y: How can Runtime Generative tools improve the Accessibility of a Virtual 3D Scene?
Cao, X., Ju, K. P., Li, C., & Jain, D. (2025) · CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) · doi:10.1145/3706599.3720265 · Cited by 3
Demo of RAVEN: Realtime Accessibility in Virtual ENvironments for Blind and Low-Vision People
Cao, X., Ju, K. P., Li, C., Potluri, V., & Jain, D. (2025) · International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS) · doi:10.1145/3663547.3759725 · Cited by 3
“Prior work has leveraged LLMs for accessibility tasks ranging from querying visual information [1] to runtime scene editing [14].”
Editing Reality: Designing In-Situ Co-Creation with Generative AI in Mixed Reality
Weng, S. C., Ma, S.-Y., Reinig, S., Kadam, P., Zhao, A. Y., Banic, A. U., Suzuki, R., & Yi–Luen, E. (2026) · Designing Interactive Systems Conference · doi:10.1145/3800645.3813087 · Cited by 3
“Complementary strands target interactive behavior generation [8, 19, 25], hierarchical co-creation [14], and prompt-based 3D/texture generation [27, 45, 46].”
RAVEN: Realtime Accessibility in Virtual ENvironments for Blind and Low-Vision People
Cao, X., Ju, K. P., Li, C., Potluri, V., & Jain, D. (2026) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3791616 · Cited by 3
Rahman, A., Yu, A., & Cho, K. (2026) · Systems · doi:10.3390/systems14020175 · Cited by 2
AI-Driven Dynamic Video Game Battles: Case Study of the Design of Tripo Clash
Wang, Z., Zheng, Y., Zhang, M., Cai, T., Huang, Z., Zhao, R., Chen, J., & Huang, X. (2025) · International Conference on Computer Supported Cooperative Work in Design (CSCWD) · doi:10.1109/cscwd64889.2025.11033545 · Cited by 2
Carcangiu, A., Manca, M., Mereu, J., Santoro, C., Simeoli, L., & Spano, L. D. (2026) · Behaviour and Information Technology · doi:10.1080/0144929x.2026.2714168 · Cited by 2
Sketch2Play: Runtime Generation of Playable 3D Game Content from Sketches
Zang, T., Chen, K. C., Li, H., & Sun, G. (2025) · ACM Symposium on User Interface Software and Technology (UIST) · doi:10.1145/3746058.3758402 · Cited by 1
GenAssist: Interactive Prompt-Driven XR Program Generation
Srinidhi, S., Singh, A., Lu, E., & Rowe, A. (2026) · IEEE Conference on Virtual Reality and 3D User Interfaces (VR) · doi:10.1109/vr67842.2026.00081 · Cited by 2
“Other applications explore this in the context of video games by dynamically generating in-game objects using text or player inputs [18, 34].”
Lottery and Sprint Arcade: Enabling Player-Driven Game Editing with Generative AI
Torii, M. G., Murakami, T., & Ochiai, Y. (2026) · The Journal of the Society for Art and Science · doi:10.3756/artsci.25.12_1 · Cited by 1
“Jennings et al. [12] introduced GROMIT, a run-time behavior generation system for Unity that compiles LLM-generated game behaviors during play without developer intervention, achieving an 85% success rate in a controlled evaluation.”
Towards Novel Gesture Recognition for Real-Time Code Generation in Extended Reality
Heinbaugh, C. E., & Kristensson, P. O. (2025) · IEEE International Symposium on Mixed and Augmented Reality Adjunct (ISMAR-Adjunct) · doi:10.1109/ismar-adjunct68609.2025.00068
“Via built-in Unity functions, the scene hierarchy is extracted and passed in a JSON format to the model for additional context when predicting a gesture, similar to the methodology used by Jennings et al. [4] and Masson et al. [16].”
InfluentialOzLand: Exploring Dynamic Generative Game Design with Multi-level Runtime Authorization
Kim, M., & Lee, Y. (2025) · ACM Symposium on User Interface Software and Technology (UIST) · doi:10.1145/3746058.3758392
Vachha, C., Kang, Y., Dive, Z., Chidambaram, A., Gupta, A., Jun, E., & Hartmann, B. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.20129 · Cited by 1
“These developments have indicated the importance of object-centric labeling and editing in the systems that are built and have guided our design of editing components of 3D scenes in VR.”
Self-citationInfiniteCATs: Semantic Crafting as a Substrate for Creative Activity Tracing
Jennings, N. B., Almeda, S. G., Hartmann, B., & Kreminski, M. (2026) · Creativity and Cognition · doi:10.1145/3803784.3816855
Self-citationMultiVerse: A Creator-Centered Approach to Steering Context-Adaptive Lyrics
Wang, A., Donahue, C., & Lindlbauer, D. (2026) · ACM Symposium on User Interface Software and Technology (UIST) · doi:10.1145/3830398.3830530
“Prior work has further explored systems that generate narrative content and gameplay behavior at runtime [19, 21, 37], moving beyond fixed branching structures.”
Torii, M. G., Murakami, T., & Ochiai, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.26447
Yang, Y., Lin, F., Xu, K., & Zhou, Z. (2026) · AIS Electronic Library (AISeL)
Tackling Open Set Gesture and Voice Interaction in 3D with Runtime Code Generation
Heinbaugh, C. E., Song, Z., Dudley, J. J., & Kristensson, P. O. (2026) · Proceedings - International Symposium on Mixed and Augmented Reality, ISMAR · doi:10.1109/ismar70079.2026.00075
Metadata
- Authors
- Jennings, N., Wang, H., Li, I., Smith, J., & Hartmann, B.
- Affiliations
- University of California, Berkeley (US)
- EECS, UC Berkeley, United States
- Venue
- Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology
- Publisher
- ACM
- Published
- 2024-10-11
- Pages
- 1-13
- OpenAlex
- W4403334210
- Cited by
- 20
- References
- 19
- Citations by year
- 2026: 8 · 2025: 12
- FWCI
- 5.77
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- ACM Symposium on User Interface Software and Technology (UIST) (publishedVersion, cc-by-sa)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence in Games
- Social Sciences › Sociology and Political Science › Digital Games and Media
- Physical Sciences › Computer Science › Information Systems › Software Engineering Research
- Keywords
- large language models, procedural content generation, game development, unity, developer workflows, gameplay
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.