OccSim: Multi-kilometer Simulation with Long-horizon Occupancy World Models

Submission Under Review

Tianran Liu    Shengwen Zhao    Mozhgan Pourkeshavarz    Weican Li    Nicholas Rhinehart
Learning, Embodied Autonomy, and Forecasting (LEAF) Lab, University of Toronto

       

---

📢 News

• [Upcoming] The source code, pre-trained models, and generation pipelines will be officially released upon the paper's acceptance. Stay tuned!

🚀 Overview

OccSim is the first occupancy world model-driven simulation via autoregressive rollouts. By eliminating the reliance on labeled HD maps during simulation, OccSim enables scalable autonomous driving simulation and achieves multi-kilometer, long-horizon generation with SoTA fidelity and diversity.

Note: We are currently organizing the codebase. Please check back later for the full release!

🔗 Citation

If you find our work interesting or helpful, please consider citing:

@article{liu2026occsim,
  title={OccSim: Multi-kilometer Simulation with Long-horizon Occupancy World Models},
  author={Liu, Tianran and Zhao, Shengwen and Pourkeshavarz, Mozhgan and Li, Weican and Rhinehart, Nicholas},
  journal={arXiv preprint arXiv:2603.28887},
  year={2026}
}