City of Light (COL)

A high-performance, Unity-based digital twin of Paris for embodied AI, robotics, and XR research.

<p align="center"> <a href="https://colab.research.google.com/drive/1Qw0uaRGRiITS5r77zU9NpuRp80KHVduO?usp=sharing" target="_blank" rel="noopener noreferrer"> <img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Colab demo"> </a>&nbsp;&nbsp; <a href="https://doi.org/10.1609/aaai.v40i48.42379" target="_blank" rel="noopener noreferrer"> <img src="https://img.shields.io/badge/Paper-AAAI%202026-orange.svg" alt="Paper"> </a>&nbsp;&nbsp; <a href="https://www.youtube.com/watch?v=KhIO3J9oGr8" target="_blank" rel="noopener noreferrer"> <img src="https://img.shields.io/badge/Trailer-YouTube-red?logo=youtube&logoColor=white" alt="Trailer"> </a>&nbsp;&nbsp; <a href="LICENSE.md" target="_blank" rel="noopener noreferrer"> <img src="https://img.shields.io/badge/Code-Apache%202.0-blue.svg" alt="Code License"> </a>&nbsp;&nbsp; <a href="LICENSE_ASSETS.txt" target="_blank" rel="noopener noreferrer"> <img src="https://img.shields.io/badge/Assets-CC%20BY--NC%204.0-purple.svg" alt="Assets License"> </a> </p> <!-- [](https://arxiv.org/abs/XXXX.XXXXX) --> <div align="center"> <strong> [ <a href="#-quick-start">Quick Start</a> · <a href="#-features">Features</a> · <a href="#-releases">Releases</a> · <a href="#-licensing">Licensing</a> · <a href="#-citation">Citation</a> · <a href="#-contributors">Contributors</a> ] </strong> </div>

---

<p align="center"> <img src="docs/images/teaser.gif" alt="COL teaser" width="820"> </p>

City of Light (COL) is a geo-anchored, city-scale simulator of Paris (~116 km²) with synchronized multi-sensor streams (RGB, Depth, Normals, Semantics) and a zero-copy Python bridge (TURBO) that sustains very high throughput (up to ~1300 FPS on a 4090 in our tests).
COL is designed for fast scripting, large-scale data collection, RL, sim-to-real and embodied research.

This repository contains both the COL build releases and PyCol, a lightweight Python stack that lets you control and interact with COL easily.

---

🧩 Features

• Geo-faithful Paris digital twin — per-tile meshes from public GIS.
• Synchronized multi-sensors — RGB / Depth / Normals / Semantics per frame.
• TURBO zero-copy bridge — shared-memory streaming to NumPy (no gRPC, no per-pixel copies).
• High throughput — frame-accurate control & observation at hundreds to thousands of FPS (resolution-scalable).
• Dynamic runtime — stochastic pedestrians & vehicles; chunk streaming with a 3×3 tile window.
• Python-first workflow — simple APIs to launch Unity, move/rotate the agent, step actions, and read frames.
• Reproducible I/O — deterministic stepping and per-frame update index.

---

🛠 Quick Start

1) Download & run COL (Unity build)

1. Download the latest build from the v0.2.0 release page:
   https://github.com/iliassarbout/CityOfLight/releases/tag/0.2.0

2. Extract the archive and run the executable for your OS.

2) Control COL from Python (PyCol)

Clone the repo and launch the demo notebook to control COL (stepping, actions, multi-sensor frames):

    git clone https://github.com/Paris-COL/CityOfLight.git
    cd CityOfLight

Then open the demo notebook and start interacting with the simulator.

---

🚀 Releases

Current public releases contain 108 tiles of 10 000 m² in the center of Paris.

---

📦 Documentation

Coming soon.

---

📜 Licensing

• Code: released under the Apache 2.0 license. See LICENSE.md.
• Assets (3D meshes, textures, etc.): released under CC BY-NC 4.0. See LICENSE_ASSETS.txt.

---

✏️ Citation

If you use City of Light (COL) in your research, please cite:

Sarbout, I.; Ounissi, M.; Cazenave-Coupet, T.; Milea, D.; and Racoceanu, D. 2026.
City of Light (COL): A City-Scale, Geo-Anchored Urban Simulator with High-Throughput Multi-Sensor Streams.
In Proceedings of the AAAI Conference on Artificial Intelligence, 40(48): 41679–41681.
https://doi.org/10.1609/aaai.v40i48.42379

BibTeX

@inproceedings{sarbout2026col,
  title     = {City of Light (COL): A City-Scale, Geo-Anchored Urban Simulator with High-Throughput Multi-Sensor Streams},
  author    = {Sarbout, Ilias and Ounissi, Mehdi and Cazenave-Coupet, Th{\'e}o and Milea, Dan and Racoceanu, Daniel},
  booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
  volume    = {40},
  number    = {48},
  pages     = {41679--41681},
  year      = {2026},
  doi       = {10.1609/aaai.v40i48.42379}
}

---

👥 Contributors & Contact

City of Light (COL) was developed between 2024 and 2025 by Ilias Sarbout. Theo Cazenave-Coupet contributed to COL during his internship (pedestrians system). Mehdi Ounissi contributed to COL through guidance on design system choices.

For questions about COL or collaborations, contact the maintainers at ilias.sarbout [at] gmail [dot] com.