Video games and reinforcement learning (RL) have a rich and intertwined history. Video games have driven major advances in RL through challenging environments such as Atari, Dota 2, StarCraft, and Minecraft. In turn, RL has fueled innovation in game development, including adaptive non-player characters (NPCs), automated quality assurance (QA), and procedural content generation.

Despite these advances, RL remains sparsely adopted in modern video game development, in part due to its reliance on massive computational resources, labor-intensive domain expertise, and hand-crafted reward functions that do not easily generalize to new tasks or genres. At the same time, the research community still lacks clear answers to fundamental challenges around sample efficiency, objective alignment, and practical deployment at scale. The second edition of the RLVG workshop looks to the future, and seeks to close this gap by investigating how video games can function as verifiable domain benchmarks for RL research, and how emerging RL methods can be translated into usable tools for contemporary video game production.

We invite submissions on topics including (but not limited to):

• Data Efficiency in Games: sample-efficient, offline, and dataset-driven reinforcement learning from gameplay data;

• Open-ended Learning for Games: continual and lifelong reinforcement learning in video games;

• World Models for Games: model-based reinforcement learning and learned world dynamics in video game engines;

• Foundation Models for Games: multimodal, generative, and language-conditioned agents for game playing and development;

• Alignment and Evaluation in Games: reward learning, preference learning, human-in-the-loop learning and evaluation frameworks that assess agent behavior with respect to human, designer, and player intent;

• Multi-agent Systems in Games: self-play, population-based training, and emergent behavior in competitive and cooperative settings;

• Benchmarks and Real-world Deployment: design of reinforcement learning benchmarks derived from modern video games, and translation of research methods into practical tools for NPC control, quality assurance, and procedural content generation.

By bringing together researchers who use video games as platforms for RL and practitioners who integrate RL into game development, this workshop seeks to surface fundamental limitations in current approaches, and outline future directions through benchmarks and methodologies that address both theoretical foundations and practical deployment challenges.

Call for Paper: 20 April

Workshop Paper Submission: 20 May

Notification of Results: 05 June

Camera-ready Submission: 29 June

Workshop Day: 15 August

All times are 23:59 PM, AoE.

Interested in sponsoring our workshop?

More information here

Email: rlvg.workshop at gmail dot com