RoboVista — Project Page

Project page for RoboVista: Evaluating Vision-Language Models for Diverse Robot Applications.

RoboVista is an expert-annotated, robot-centric Visual Question Answering benchmark built with the modular Robot Question Answering (RQA) framework — 474 multiple-choice questions across 6 robot application domains (agriculture, industrial, domestic, surgical, autonomous driving, and open robot datasets) and 39 task types.

Stack

• Vite + React 18
• No CSS framework — a single hand-written src/index.css
• All page copy and data live in src/content.js (single source of truth)

Develop

npm install
npm run dev      # http://localhost:5173/robovista/

Build & preview

npm run build
npm run preview

Deploy (GitHub Pages)

The site is configured for https://berkeleyautomation.github.io/robovista/ (base: '/robovista/' in vite.config.js, homepage in package.json).

npm run deploy   # builds and pushes dist/ via gh-pages

Editing content

Edit src/content.js — every section (hero, TL;DR, RQA framework, benchmark, construction, evaluation tables, physical-robot validation, findings, future work) reads its text and numbers from there. Figures live in public/static/images/ and are referenced through the asset() helper so they resolve under the project base path.

Structure

src/
  App.jsx                 # section order
  content.js              # all copy + data
  index.css               # styles + theme tokens
  components/
    Hero.jsx              # title, authors, links
    RoboVistaLogo.jsx     # inline SVG wordmark
    Teaser.jsx            # overview figure (+ shared Caption)
    Tldr.jsx
    RQAFramework.jsx      # modular abstraction + module tuple
    Benchmark.jsx         # domains, gallery, examples, stats
    Method.jsx            # 3-stage curation + case studies
    Results.jsx           # zero-shot / CoT / ICL tables
    Physical.jsx          # bimanual alignment + surgical knot tying
    KeyFindings.jsx
    FutureWork.jsx
    Footer.jsx
    SideNav.jsx
    Icons.jsx