Papers for

real-time graphics programmers

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

HyperDrone enables ultra-fast realistic drone simulation on GPUs

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

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.

Tue 20 OctRoboticsArtificial IntelligenceMachine Learning
The gist
Simulating drones and their cameras realistically is usually slow, making it hard to train real-world flying skills in a virtual setting. The authors created HyperDrone, a system that runs thousands or millions of drone simulation frames per second by doing physics and advanced 3D rendering entirely on graphics cards. This speed lets them train a quadrotor to control its rotation using only camera views in simulation and then make it work on a real tiny drone with its own camera. Their system makes realistic drone simulation much faster and practical for real robot training.
Open → 2610.28950v1