Open Problems from Industry

IEEE CoG 2026 · Industry Day · Session 2

Identifying Research Gaps and Publishing Industry Perspectives

This session creates a structured forum for identifying open problems, research gaps, and opportunities for collaboration within the games research ecosystem, with a focus on surfacing challenges that are visible in industry but underrepresented in academic conversation.

If you are interested in attending this workshop, please complete the Open-Problems Proposal form below. Our workshop Discord server is also coming soon! 

If you are interested in presenting the problem as one of our lightning talk speakers, please check the box in the form. 

All submissions are DUE Friday, August 14, 2026 11:59:59 PM AoE.  

A Google Doc version of the ToG@CoG Session calls is available HERE. 

Schedule 

Friday September 4, 2026 (Industry Day) Madrid Local Time

14:15 – 14:50

Lightning Talks 35 min

14:50 – 15:05

Collaborative Breakout Discussions 15 min

15:05 – 15:15

Cross-group Discussion and Synthesis 10 min

Expected Outcome

• A stronger community of academic and industry researchers

• Continued discussion spaces and working groups (e.g., Discord) 

• A community-driven roadmap, editorial, or open-letter publication to be published in IEEE Transactions on Games (ToG)

• Strategic plans to foster future partnerships 

• Actionable recommendations for conferences, journals, and publication models

Accepted Open Problems

When Does Game Assistance Start Replacing the Player’s Strategy?

Speaker: Alessandro Penno

Abstract: Tom and I first worked together around Old School RuneScape. I was the product manager for the game client and plugins, while Tom built Quest Helper, the RuneLite plugin that guides players through quests. That experience showed us how useful timely assistance can be. It also raised a harder question: when does help stop preserving the player’s strategy and start supplying it? The talk proposes a simple comparison when a player is stuck: silence, a question that prompts reflection, or a direct hint. The test is whether the player later completes a similar task without help. The open problem is how games can assist players while preserving their agency, and how researchers can distinguish support from strategy substitution. 

The Expansion of Games as a Service models 

Speaker: Sergio Gutiérrez Manjón

Abstract: The problem lies in how the dominance of Games as a Service and free-to-play models cannibalizes traditional video game production. Offering endless, free content traps players inside single digital ecosystems driven purely by financial gain, stifling game variety. Within these platforms, aggressive dark patterns and persuasive architectures target youth, using manipulative behavioral loops to maximize screen time, normalize micropayments, and prompt compulsive spending. 

Addressing this phenomenon is crucial now because digital games define youth socialization, making young players with developing cognitive self-regulation vulnerable to dark designs. Mechanics like loot boxes closely mirror gambling, triggering predatory monetization and social pressure that blur the line between play and economic exploitation. 

Solving this issue remains extremely difficult due to lagging institutional oversight. Although updated frameworks like PEGI attempt to raise age ratings for predatory design, enforcement remains insufficient while regulatory bodies struggle to keep pace with rapidly evolving monetization tactics. 

Can We Make 3D Reconstruction Truly Reproducible for Indie Developers?

Speaker: Gyanendra Vyas

Abstract: One challenge I have encountered while building a production workflow around 3D Gaussian Splatting is that the technology is advancing much faster than the tooling and workflows needed to make it genuinely accessible and reproducible. As an independent game developer in India, I have been exploring how to take a fully authored Unreal Engine environment, reconstruct it as a Gaussian Splat, optimize it, and publish it directly to the web. While the individual components exist, putting them together into a reliable, repeatable pipeline still requires significant technical knowledge, experimentation, compute resources, and fragmented tooling. 

This matters because Gaussian Splatting has the potential to make high-quality 3D experiences useful far beyond research labs, including games, education, XR, digital heritage, and architectural visualization. However, developers with limited resources can struggle to reproduce published results or understand which capture, reconstruction, optimization, and rendering choices actually matter. This creates a gap between research prototypes and practical industry adoption. 

The problem is particularly visible from emerging game-development ecosystems such as India, where small studios, students, and independent creators often do not have access to expensive scanning hardware, dedicated research infrastructure, or proprietary tools. My work has shown that an Unreal Engine-to-Gaussian-Splat-to-Web workflow is possible using largely open-source tooling, but it has also exposed unanswered questions around automated capture coverage, reproducibility, quality evaluation, cross-device performance, and standardized workflows. 

I believe there is an opportunity for the graphics community and industry to work together on an open, reproducible pipeline and shared benchmarks for this problem. Rather than simply consuming research from the Gaussian Splatting community, indie developers and practitioners should also be able to contribute real-world production findings back to it. I would like to use this session to discuss what such a community-driven workflow could look like and identify the research problems that need to be solved to make it practical.

Supporting Blind and Low-Vision Players

Speaker: Khaloud Al-Buainain‏ 

Abstract: Multiple guidelines exist to support developers in building accessible gaming experiences, including the Game Accessibility Guidelines, Accessible Player Experiences (APX), and Xbox Accessibility Guidelines. Despite the growing field of game accessibility, there remains a significant gap in accessibility support for Blind and low-vision (BLV) players. Research and guidance on game accessibility for BLV players is fragmented, often buried across general game accessibility guidelines, player communities, and academic research. The number of games that provide comprehensive accessibility for BLV players remains limited, with accessible implementations often concentrated within particular games or genres. Current research identifies many of the elements that should be made accessible, but provides limited guidance on how to translate visually intensive gameplay into a non-visual experience. Indie developers and developers without accessibility expertise may therefore struggle to translate broad recommendations into meaningful and practical design decisions. 

One example is the translation of visual information into auditory information, which forms a major component of game accessibility for BLV players. However, existing guidance does not always address in sufficient depth how developers should manage audio hierarchy, audio ducking, cognitive overload, distinguishable auditory cues, or the prioritization of competing audio information. It is not enough to simply translate visual information into sound; developers must determine which information should be communicated, when it should be communicated, how it should be distinguished from other sounds, and how much information a player can reasonably process at once. 

The limited number of Blind-accessible games highlights a research-to-industry gap. There exists knowledge surrounding the experiences, barriers, and needs of BLV players is growing, yet there remains a lack of detailed, actionable guidance on how developers can translate this knowledge into accessible game design and implementation. 

This problem is particularly important now as more mainstream games are beginning to support BLV players, with a prominent example being The Last of Us Part II (TLOU2) by Naughty Dog. The release of TLOU2 demonstrated that a complex, visually intensive mainstream game can provide an extensive non-visual gameplay while maintaining meaningful player agency. The game provides an important benchmark and a concrete example of how BLV gamers can be supported. However, given the specific genre, mechanics, and design of TLOU2, its accessibility solutions cannot simply be transferred across all games. 

Different game genres vary substantially in their mechanics, interactions, goals, environments, and information demands. In addition, BLV players are diverse in their abilities, levels of experience, preferences, and use of assistive technologies. Therefore, there cannot be a single accessibility solution that applies to every player or every game. Furthermore, accessibility features interact with other elements of game design, including challenge, immersion, fairness, and player agency. Finding the balance between making a game accessible while preserving an engaging and meaningful gameplay experience is therefore difficult and requires close collaboration between BLV players, game developers, and accessibility researchers. 

The challenge is not simply identifying accessibility barriers. It is determining which solutions work, how they work, in which contexts they work, why they work, and how they can be implemented while preserving the true essence of the game. While these accessibility barriers are difficult to overcome, it is still not game over.

Generalist AI Agents in Video Games

Speaker: Iuri Frosio

Abstract: What is the problem? We have found that many generalist AI agents that perform well on real world data have limited understanding of the video game scenarios. This fact limits the application of AI in the realm of video games; at the same time, equipping generalist AI agents with such capability turns out to be a non-trivial problem for the several reasons illustrated in the following. 

Why is it important now? The “dream” of many gaming companies is an AI agent that is capable not only of zero-shot playing any (novel) game, but also of understanding the gaming scenario from a point of view that is similar to that of a human player. When equipped with full understanding of the game visuals and logic, an AI agent can then not only play autonomously or following the supervisor instructions, but also inspect the game in search of bugs; supervise and comment a sequence of human player actions; replicate a gaming sequence following the engineer instructions for benchmarking; or report back to the game creator after finding gaming sections that are too easy or difficult to play, thus suggesting how to improve the game design for a better player experience. Overall, such AI agents would allow development studios to spend more time focusing on game concepts, visuals and logic to make sure the game is fun, rather than spending as much time finding and fixing bugs, testing gaming scenarios, and so on. 

Why is it hard to solve? There are several reasons explaining why achieving such a goal is hard and many of them are (or require solutions that are) entangled with each other. These include: 

• 1 Video game physics and logic are different: the physical rules that apply in video games are often a super set of real world physics. For instance, in video games the strength of the gravity force is often different from that on Earth, players can sometimes cross or see-through objects that are solid and not-transparent in the real world, or vehicles can crash without reporting damages. At the same time, the logic rules followed in video games can be counterintuitive: for instance, doors can be opened using a fruit, vehicles can be repaired by crashing into barriers at the right speed, or actions in a given an unusual sequence must be performed to gain access to the next level. Since real world agents are not trained under these conditions, they hardly generalize (even when prompted with gaming instructions). The natural solution would be to re-train from scratch (or fine tune) these agents on gaming data, but…

• 2 Large and complete video games dataset are not available: in fact, there is nothing like a free-to-use, large dataset of video game data that can be used for training or fine tuning AI agents. Some datasets are either small and old (like those used to train on Atari games), while the best gaming agents (including world models in this case as well) published so far (such as Genie3, Sima2, Optimus3) are either trained on non public datasets or on specific games. One of the reasons for this is that…

• 3 Use of video game data for training is subject to limitations: video game data are often IP protected and their legal use in scenarios different from simple playing is forbidden or in lie in a legal grey area, complicating the life of researchers. As an alternative solution for the development of effective (and potentially even more computationally efficient) agents that interact with videogames, one could think of allowing them to access the internal game state, but…

• 4 The internal state of video games is not accessible: in fact, beyond IP issues (the video game code is most of the time proprietary and not available), allowing access to video game information would also open the door for cheaters, making the entire video game system less safe. A last resort would be then to… 

• 5 Develop your own gaming scenarios: these should be general enough to allow agents to generalize to real games: huge augmentation in the game logics, visual and physics would be required to guarantee generalization, but this also implies that the trained agent would be large and computationally intense, making its application to real case problematic. And even in this case, the last but not least problem would remain that…

• 6 Benchmarking AI performance remains an open problem: in fact, without a well established and commonly accepted set of benchmarks, the quantification of the performance and thus also the comparison of different models remain not well defined, if not when resorting the base metrics (like the number of kills per minute in FPS video games) that however only tell half of the story. 

We finally notice that recent developments go in the direction of training world models for video games. Although equipping a model with the capability of predicting the future is likely going to significantly improve the overall performance (likewise for real world models), this does not solve all the problems listed here: large training datasets or usage of games for training are still needed, with all the connected issues.

Game Code That Is Not Gameplay

Speaker: Ezequiel Santos

Abstract: Mobile games increasingly depend on code that has little to do with gameplay itself: telemetry and analytics, advertising and mediation, attribution, consent, crash reporting, remote configuration, authentication, payments, and other platform services. These systems are important parts of operating a commercial game. For example, research on game analytics shows how telemetry is used for retention and funnel analysis, A/B testing, soft launches, and monetization decisions. 

The problem is that these technologies are often built for the wider mobile-app ecosystem and then consumed by developers working primarily inside a game engine such as Unity or Unreal. An SDK may expose a simple engine-facing API while its integration depends on native Android and iOS libraries, Gradle or Maven, Xcode, package managers, manifests, privacy configuration, and several layers of adapters and dependencies. As a result, installing or upgrading an SDK can require developers to work outside their normal environment and understand failures originating several layers below the API they actually use. 

This matters now because commercial games increasingly depend on these services, while both game engines and native mobile platforms continue to change. The problem is also not limited to builds failing. An integration can compile successfully while telemetry is incorrectly configured, consent does not propagate through a wrapper SDK, an advertising adapter is incompatible, or behaviour changes after an SDK upgrade. When something breaks, it may be difficult to determine whether the cause is the game engine, the SDK wrapper, a native dependency, the build system, platform configuration, or another third-party component. 

The open problem is how SDK teams should design the integration experience for game developers: what complexity should be hidden, what should remain visible, and how can developers diagnose problems when the abstraction breaks? Existing research examines game analytics, mobile SDKs, cross-platform development, privacy configuration, and game-engine API usability, but these areas are largely studied separately. We need better understanding of the complete developer experience of installing, configuring, validating, upgrading, and debugging the technologies that make modern mobile games work but are not themselves gameplay.

The Drop-Off: Women in Games from Players to Power and Capital 

Speaker: Marie-Claire Isaaman

Abstract: Women make up around half of the global games player base, yet their representation falls dramatically as we move through the games industry towards positions of greater power, ownership and access to capital. 

Women in Games' 2026 insight report, The Drop-Off: Women in Games from Players to Power and Capital, brings together UK, European and global evidence to reveal a striking progression: women represent around 50% of players, approximately 25–30% of the games workforce, around 22% of senior roles and approximately 16% of executive leadership. At the point where we reach ownership, investment and capital, however, something significant happens: the data itself largely disappears. 

As The Drop-Off puts it: “Workforce participation is measured. Player demographics are measured. Market performance is measured. Access to capital is not.” 

There is currently no robust, consistently reported, games-specific dataset showing funding and investment outcomes by gender. We therefore cannot reliably answer some remarkably basic questions: Who applies for funding? Who succeeds? How much do they receive? Who receives follow-on investment? Who scales? And where in that journey do women disappear? 

This matters beyond Women in Games. The European Commission's Gender Equality Strategy 2026–2030 calls for gender equality to be embedded across economic sectors and EU policy. Its priorities include attracting more women into research, innovation and start-ups, improving access to finance for women entrepreneurs and strengthening gender-sensitive approaches to funding. The games industry is an important European creative, technological and economic sector, yet currently lacks much of the evidence needed to understand whether these ambitions are being realised within games. 

This lightning talk will therefore present The Drop-Off not as the answer, but as a research challenge to the academic games community.

There are two connected problems. The first is the data gap. How might researchers help us create robust, standardised and longitudinal ways of measuring gender, leadership, ownership and access to capital across games? What should public funders, publishers, investors and industry bodies collect? Can fragmented datasets be connected? And how can we build methods that recognise intersectionality while remaining practical enough for industry-wide adoption? 

The second is the power and capital gap. Better data can tell us where the drop-off occurs, but research is needed to understand why. How do networks, organisational structures, perceptions of risk, investor pattern recognition and access to decision-makers influence progression from developer to leader, founder and investor? Are there identifiable points where women disproportionately leave, stall or are excluded? And which interventions actually change those outcomes? 

This is where collaboration between industry and academia could be transformative. Games researchers can help move us from correlation towards explanation: developing longitudinal studies, establishing common definitions and benchmarks, interrogating funding and investment systems, comparing countries and ecosystems, and rigorously evaluating interventions. 

The report argues that “the lack of data is not a neutral limitation – it is a barrier to accountability, investment efficiency and sustained growth.” The opportunity, therefore, is bigger than producing another dataset about women in games. It is to create the research infrastructure needed to understand how opportunity, power and capital move through the games ecosystem – who gets access to them, who does not, and why. 

The Drop-Off identifies the warning signal. My challenge to the CoG research community is: Can academia and industry work together to measure the drop-off, understand what causes it, and build the evidence needed to change it?

Organizers