Special session: Evaluating and Advancing Spatial Intelligence through Games

Organizing committee

• Prashant Jayannavar — University of Illinois Urbana-Champaign, US (paj3@illinois.edu)
• Alessandro Suglia — University of Edinburgh, United Kingdom (asuglia@ed.ac.uk)
• Sina Zarrieß — University of Bielefeld, Germany
• Massimo Poesio — Queen Mary University of London, United Kingdom

Scope

Spatial intelligence, the ability to perceive, reason about, and manipulate spatial relationships, is fundamental to human cognition and essential for artificial intelligence systems operating in both physical and virtual environments. Games provide rich, controlled, and interactive testbeds for evaluating and advancing spatial intelligence in AI, offering diverse scenarios that require understanding spatial configurations, navigation, object manipulation, and communication about spatial concepts.

This special session seeks contributions that use games as diagnostic tools or environments for developing and evaluating spatial intelligence in AI systems. We welcome research spanning both embodied and unembodied games, as well as multimodal and unimodal settings. Of particular interest are collaborative and interactive scenarios where spatial intelligence enables AI agents to serve as instruction followers or instruction givers, playing alongside or assisting humans in open-ended gameplay or task-specific applications such as navigation, construction/assembly, object manipulation, etc. Further, reward modeling in such domains also inherently requires spatial understanding, and we encourage work that studies this.

The session is in part motivated by prior workshops such as the 4th Workshop on Spatial Language Understanding and Grounded Communication for Robotics (SpLU-RoboNLP 2024), When Language meets Games Workshop (Wordplay 2025), with particular inspiration drawn from the Dagstuhl Perspectives Workshop Human in the Loop Learning through Grounded Interaction in Games (https://www.dagstuhl.de/seminars/seminar-calendar/seminar-details/24492).

Topics of Interest

We are seeking papers that present methodological contributions to relevant tasks as follows.

Relevant Tasks

Including, but not limited to:

• Core instruction following and giving tasks (dialogue-based or single-turn)
• Related sub-problems, e.g., referring expression comprehension/generation, clarification question generation, planning
• Reward modeling
• Novel task formulations to evaluate spatial intelligence capabilities

Methodological Contributions

For the above tasks, we invite contributions including, but not limited to:

Data and Resources

• Data collection, synthetic data generation, and simulation frameworks
• Data scarcity in embodied or interactive settings
• Inverse Dynamics Models and related approaches for pseudo-labeling, enabling scalable dataset creation from abundant unlabeled sources
• Resources and environments for interactive/online learning

Modeling Approaches

• LLMs, VLMs, VLAs, agentic frameworks
• Reinforcement Learning
• Model design, fine-tuning strategies, in-context learning, parameter-efficient methods, and specific techniques for spatial reasoning or spatio-temporal memory representations

Evaluation and Analysis

• Automated metrics, human evaluation studies, and benchmark design
• Games as diagnostic environments for behavioral analysis (manual or automatic) of models to uncover their strengths and limitations
• Evaluating reward modeling ability itself as an effective proxy for spatial reasoning ability

Program Committee members

• Julia Hockenmaier — University of Illinois Urbana-Champaign
• Marc-Alexandre Cote — Microsoft Research – Montreal
• Raffaella Bernardi — Free University of Bozen-Bolzano
• David Schlangen — University of Potsdam
• Manling Li — Northwestern University
• Parisa Kordjamshidi — Michigan State University
• Simon Dobnik — University of Gothenburg
• Nikolai Ilinykh — University of Gothenburg
• Vardhan Dongre — University of Illinois Urbana-Champaign
• Ruiyi Wang — University of California San Diego
• Sandro Pezzelle - University of Amsterdam

Submission Instructions

We invite the submission of full *technical* papers with an 8 page limit. Full *technical* papers should provide a technical or empirical contribution to scientific or engineering aspects of games. Relevant dates for this call are as follows:

• Submission of full technical papers: 17th March 2026
• Notification of acceptance of full technical papers: 1st May 2026
• Submission of the camera-ready version of full technical papers: 28th May 2026

All page limits include references and appendices.
NONE OF THE SUBMISSION DEADLINES WILL BE EXTENDED.
All deadlines are Anytime on Earth (AoE).

Papers must be submitted through the conference submission system available at the following link: https://easychair.org/conferences?conf=ieeecog2026

All paper submissions should follow the recommended IEEE conference author guidelines. MS Word and LaTeX templates can be found at https://www.ieee.org/conferences/publishing/templates

All submitted papers will be fully peer-reviewed, and accepted papers will be published in the conference proceedings and on IEEE Xplore. CoG will use a *double-anonymous review process*. Authors must omit their names and affiliations from their submissions, avoiding obvious identifying statements. Submissions not abiding by anonymity requirements will be desk rejected.

Papers might be allocated to either poster presentations or oral presentations.