Paper on arXiv | Live demo (browser) | Documentation

Requirements

Ubuntu

apt update
apt install git cmake curl python3-dev python3-pip python3-venv build-essential

• Nvidia GPUs / CUDA: Install CUDA toolkit and set environtment variable CUDACXX={cuda_toolkit_path}/bin/nvcc
• Other GPUs / Vulkan: apt install libvulkan-dev glslang-tools

macOS

brew install python cmake

Quick Start

pip install hyperdrone[examples]

Benchmark your GPU:

python3 -m hyperlux.examples.benchmark

HyperDrone Spy is the UI that lets you peek into HyperDrone environments. Here, 4 agents share a scene (for perf) but don't see each other (e.g. for single-agent RL). HyperDrone also supports flexible configurations (e.g. multi-agent without self-occlusion).

python3 -m hyperdrone.examples.spy

Documentation

Please check out the documentation at docs.hyperdrone.dev

Citing

When using our code/setup in an academic work please cite our publication using the following Bibtex citation:

@misc{eschmann2026hyperdrone,
      title={HyperDrone: Simulation-to-Reality Transfer at Millions of Frames per Second}, 
      author={Jonas Eschmann and Giuseppe Loianno},
      year={2026},
      eprint={2610.28950},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2610.28950}, 
}