VectorWorld enables stable, real-time 1km+ closed-loop world model rollouts for autonomous driving using diffusion flow on vector graphs.
arXiv · March 19, 2026 · 2603.17652
The Takeaway
It solves the 'compounding error' problem in AV simulation by using a motion-aware gated VAE and kinematic logit shaping. This allows for long-horizon interactive simulation that is physics-consistent, a critical requirement for evaluating end-to-end driving policies.
From the abstract
Closed-loop evaluation of autonomous-driving policies requires interactive simulation beyond log replay. However, existing generative world models often degrade in closed loop due to (i) history-free initialization that mismatches policy inputs, (ii) multi-step sampling latency that violates real-time budgets, and (iii) compounding kinematic infeasibility over long horizons. We propose VectorWorld, a streaming world model that incrementally generates ego-centric $64 \mathrm{m}\times 64\mathrm{m}