HyperDrone: Simulation-to-Reality Transfer at Millions of Frames per Second

Jonas Eschmann, Giuseppe Loianno

University of California, Berkeley

Published October 20, 2026 · on the feed October 23, 2026 · ▲ 11 · arXiv 2610.28950

The abstract

We present HyperDrone, an extremely fast quadrotor simulator focused on high-performance batch rendering that is more than 20x faster than existing state-of-the-art batch quadrotor simulators. HyperDrone combines highly optimized multirotor dynamics with HyperLux, a new ray-tracing engine that supports Nvidia OptiX, Vulkan, Apple Metal and WebGPU backends that takes full advantage of modern ray-tracing hardware, capable of running both components entirely on the GPU. HyperLux renders RGB, depth, semantic/instance segmentation, surface normals and optical flow observations at configurable levels of fidelity, including reflections, realistic lighting, motion blur, anti-aliasing, and Physically Based Rendering (PBR). Even in the highest-fidelity settings, HyperLux achieves millions of Frames Per Second (FPS) on consumer-grade GPUs while rendering realistic 3D scenes. We demonstrate the merits of HyperDrone by training a vision-based yaw-control policy purely in simulation and transferring it to a real small-scale quadrotor (Crazyflie) with a camera, running completely onboard.

The Brief

2:02

A short podcast summary of the paper. Written and voiced by AI.

Transcript

HyperDrone. Simulation-to-Reality Transfer at Millions of Frames per Second. By Jonas Eschmann and one other, from the University of California, Berkeley. Published on the twentieth of October. Its code repository is public but slight: three source files.

Training a quadrotor from camera images in simulation depends on rendering those images in enormous batches, and speed is the constraint this work goes after. Existing batch quadrotor simulators set the bar it measures itself against. The test of any such simulator is twofold: it has to produce observations quickly enough to train on, and realistic enough that what is learned from them still works once a real vehicle leaves the screen and takes to the air.

HyperDrone is a quadrotor simulator built around high-performance batch rendering, and it reports being more than twenty times faster than existing state-of-the-art batch quadrotor simulators. It pairs highly optimized multirotor dynamics with a new ray-tracing engine called HyperLux, and both components can run entirely on the GPU.

HyperLux is designed to take full advantage of modern ray-tracing hardware, with backends for Nvidia OptiX, Vulkan, Apple Metal and WebGPU. It renders RGB, depth, semantic and instance segmentation, surface normals and optical flow. Fidelity is configurable, and at the high end includes reflections, realistic lighting, motion blur, anti-aliasing, and physically based rendering.

Even at its highest-fidelity settings, HyperLux reaches millions of frames per second on consumer-grade GPUs while rendering realistic 3D scenes. To show the simulator is useful and not only fast, the authors train a vision-based yaw control policy purely in simulation and transfer it to a real small quadrotor, a Crazyflie with a camera, where it runs completely onboard.

The claim in the title is the pairing. Millions of frames per second on ordinary hardware is the speed half, and a policy that leaves simulation and flies a real drone using its own onboard compute is the transfer half. HyperDrone presents them as one package, with rendering fidelity left as a setting the user can turn up or down.

More with an account

Listen to every day’s papers

Daily playlists, deep dives, quests for the papers, and quizzes on their key terms.

Join free
KipQuest: the day’s AI research, as podcasts and quests. All papers Android app Privacy Contact Episodes are written and voiced by AI from each paper. Summaries can contain mistakes: the paper is the source.