Agentic Game Development as a Verifiable Trajectory Data Engine for Scaling World Models

Hugging Face Daily Papers Papers
game-engine world-model reinforcement-learning rl-post-training verifiable-data human-engine-verification

Summary

This paper proposes using game engines as a verifiable trajectory data engine for scaling world models, introducing RLHEV to combine dense engine signals with human feedback for reinforcement learning post-training.

A common strategy for scaling world models is to train on more crawled video with more compute. We argue that this strategy is inefficient: scaling world models also requires a recursive data engine that offers grounded reward signals. The success of code agents illustrates why this matters. As code is executable, compilers and runtimes can provide high-quality rewards for Reinforcement Learning (RL) post-training of LLMs. By contrast, spatial generation still relies largely on fuzzy proxies such as CLIP scores. These signals are fuzzy and biased, making them hard to support RL post-training. Compared with these, game development provides a missing reward environment for spatial world models. A scene encoded by a game engine is an executable world specification: the engine can efficiently check collision, physics, navigability and bounded playability, while the developer provides the global verification signal by judging whether the scene should be accepted. Game development also provides real-world long-horizon trajectory data for RL post-training. We therefore propose Reinforcement Learning with Human-Engine Verification (RLHEV), a post-training paradigm that combines dense engine signals with implicit human acceptance feedback from the development process.
Original Article
View Cached Full Text

Cached at: 08/28/26, 07:24 AM

Paper page - Agentic Game Development as a Verifiable Trajectory Data Engine for Scaling World Models

Source: https://huggingface.co/papers/2608.25518

Abstract

Game engines provide executable verification and long-horizon trajectories for reinforcement learning post-training of spatial world models, motivating a human-engine verification paradigm.

A common strategy for scalingworld modelsis to train on more crawled video with more compute. We argue that this strategy is inefficient: scalingworld modelsalso requires a recursive data engine that offers grounded reward signals. The success of code agents illustrates why this matters. As code is executable, compilers and runtimes can provide high-quality rewards forReinforcement Learning(RL) post-training of LLMs. By contrast, spatial generation still relies largely on fuzzy proxies such asCLIP scores. These signals are fuzzy and biased, making them hard to supportRL post-training. Compared with these, game development provides a missing reward environment for spatialworld models. A scene encoded by agame engineis anexecutable world specification: the engine can efficiently checkcollision,physics,navigabilityandbounded playability, while the developer provides the global verification signal by judging whether the scene should be accepted. Game development also provides real-world long-horizon trajectory data forRL post-training. We therefore proposeReinforcement Learning with Human-Engine Verification(RLHEV), a post-training paradigm that combines dense engine signals with implicit human acceptance feedback from the development process.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2608\.25518

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2608.25518 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2608.25518 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2608.25518 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

From Pixels to States: Rethinking Interactive World Models as Game Engines

Hugging Face Daily Papers

This paper rethinks interactive world models as game engines by examining four key dimensions—action control, state dynamics, state-observation persistence, and real-time generation—and introduces a scalable data engine for Black Myth: Wukong with over 90 hours of gameplay data to advance state-aware game world modeling.

Multiplayer Interactive World Models with Representation Autoencoders

Hugging Face Daily Papers

This paper introduces MIRA, the first large-scale multiplayer world model for highly dynamic physics-based environments, trained on 10,000 hours of Rocket League gameplay. The 5-billion-parameter latent diffusion model generates stable four-player rollouts in real time, with distributional quality holding steady for hours.

Scaling Automatic Research Agents via World Models

arXiv cs.LG

This paper identifies a scalability bottleneck in RL-trained automatic research agents—environment execution dominates training cost—and proposes World Model RL (WMRL) with online debiasing and inverse-variance denoising to replace real execution, achieving 3–4x training speedups and better generalization.

Reward as An Agent for Embodied World Models

arXiv cs.AI

This paper introduces Reward as an Agent and DynDiff-GRPO to address reward hacking and limited exploration in reinforcement learning for embodied world models, achieving significant accuracy gains.