VisualPatchWorld (VPW) represents world dynamics as executable code. According to the supplied abstract, it uses short active probes to select a qualitative dynamics form, then fits free parameters from state-action traces by minimizing multi-step prediction error. The resulting program supports simulation, inspection, editing, and model-predictive control, with scene graphs providing updated state. The abstract reports 69.0% mean planning success, 23.5 percentage points above the strongest code baseline. However, the arXiv identifier and publication date point to July 2026, so the paper and results cannot currently be independently verified from the supplied metadata alone.
No heat snapshots are available in the last 24 hours.