BoxWorld separates interactive reasoning from generative rendering in game world models
What happened
Zike Wu et al. propose BoxWorld, an interactive video world framework that separates interaction reasoning from generative rendering. User-defined interaction rules are first processed by the Box World Engine, which produces object states. The Box-State generative renderer then autoregressively renders real-time interactive video from those object states. The paper reports that BoxWorld clearly outperforms end-to-end world models on its own PairedStateBench benchmark and remains competitive on several public benchmarks. These are the paper authors' conclusions. Training has three stages. The first uses automatically annotated open-domain and game videos to learn object localization and state transitions. The second performs rollout distillation on the model's own generated histories.
AI-written from the coverage · updated 25 d ago
Coverage timeline
Follow the coverage to see the story from different sides.
- arXiv Game World Models & GenerationPickBoxWorld: Let World Engines Control and Video Generators Render
BoxWorld hands user-defined interaction rules to the Box World Engine, then has the Box-State generative renderer autoregressively render a real-time interactive video world from the object states. The paper says it clearly outperforms end-to-end world models on evaluations including PairedStateBench and remains competitive on several public benchmarks. Training has three stages: learning object localization and state transitions from automatically annotated open-domain and game videos, then rollout distillation on self-generated histories.