Valerant is a new framework presented to address the challenge of creating persistent and navigable 3D game maps. It couples predictive visual rollouts with SLAM-based spatial reconstruction and exploration-driven action selection. The framework transforms a pretrained action-conditioned world model into a WAM for exploring and constructing 3D game maps. It operates beyond 2D visual simulation, enabling interaction in a persistent 3D environment.
Existing game-oriented approaches often combine action-conditioned world models with external policies and reward functions. Valerant distinguishes itself by instantiating persistent 3D geometry, mirroring the approach used in autonomous driving and robotics. This allows selected actions to be executed within a physically persistent environment.
The framework starts with a single image and progressively transforms it into a complete 3D game map. This process utilizes SLAM-based spatial reconstruction and exploration-driven action selection. This approach offers a new approach to reducing manual effort in 3D game-map creation.
This work extends WAM-based interaction beyond 2D visual simulation. It addresses the need for navigable 3D games that require explicit geometry for movement and interaction.
Source: https://arxiv.org/abs/2609.09418