Skip to content

README.md

Latest commit

 

History

History
49 lines (32 loc) · 2.17 KB

README.md

File metadata and controls

49 lines (32 loc) · 2.17 KB
Raw

GAUGE Web Demo

An interactive research project page for GAUGE: A Measurement-Grounded Benchmark for Physical Fidelity in Simulation Engines and Video World Models.

The demo translates the paper into five explorable ideas:

  • a real-vs-simulation trajectory instrument;
  • the two complementary evaluation tracks;
  • an atlas of 22 rigid, textile, cable, and volumetric task families;
  • an interactive comparison of Isaac Sim, Genesis, and Newton results;
  • a three-layer diagnosis of video world models: equation form, parameter accuracy, and temporal stability.

The task atlas opens as a multi-video gallery so every real trial stays attached to its task, target, materials, and paper detail. A Dataset toggle provides a sortable 22-task table when a denser view is useful. The engine and world-model sections use linked tables for task, metric, scene, and prompt exploration.

Run locally

Requires Node.js >=22.13.0.

npm install
npm run dev

Open http://localhost:3000.

Validate

npm test

The test command creates a production build and verifies the rendered research page and project-specific metadata.

Project map

  • app/GaugeDemo.tsx — content, data, and interactions
  • app/globals.css — responsive visual system
  • public/paper/ — paper-derived figures and experiment crops
  • public/trials/ — short, web-optimized previews of the real-world trials
  • public/gauge.pdf — source paper used by the local demo
  • public/og.png — generated social preview card

The project uses vinext and retains .openai/hosting.json for optional future hosting, but this version is configured and verified for local use.

Task metadata and selected trial media were cross-checked against the companion NINGYURICHARD/gauge-web implementation, then reorganized for this interface.

Media strategy

The site serves compact H.264 previews so first-load and playback costs stay low. Full-resolution source captures should live in Git LFS, a GitHub Release, or object storage when they are added; the browser-facing previews stay in the regular site artifact because they are intentionally small.