Mirai is a browser-first robotics platform that lets users design robot arms, generate tasks from natural language, validate plans, run physics simulation, and export hardware-ready code.
Built for the Transforming Enterprise Through AI Hackathon (Track 3: Robotics and Simulation).
- Frontend app: https://mirai-tech-ex-hackathon-transformin.vercel.app
- Backend API: https://production.up.railway.app
- Repository: https://github.com/Mizunandayo/mirai
- Open the frontend app.
- Go to Design and load a robot preset.
- Go to Tasks and generate a plan from natural language.
- Review AI pre-flight validation status.
- Run simulation and verify playback.
- Export artifacts (Arduino, Python, BOM, URDF, signed ZIP).
- Core build complete through Day 7 milestones.
- Day 8 polish and submission tasks are in progress.
- 60fps in-browser simulation flow is implemented with deterministic motion playback.
- Dual-physics architecture is live:
- Rapier WASM in browser for real-time simulation.
- MuJoCo on backend for validation, divergence checks, and servo lifespan prediction.
- Gemini-powered planning and repair workflow is integrated in the task editor.
- Community library is implemented with seeded tasks and famous preloads.
- Export pipeline is live for Arduino, Python, BOM, URDF, and signed ZIP bundles.
- Design custom robot arms in 3D with segment and gripper configuration.
- Define tasks through a visual graph editor or natural language input.
- Generate structured plans using Gemini based on scene and arm context.
- Run deterministic safety and feasibility checks before simulation.
- Simulate tasks in-browser at interactive framerates.
- Validate execution accuracy against MuJoCo backend simulation.
- Export build-ready artifacts for real hardware workflows.
- SceneGraph
- TaskSpec
- ValidationReport
- ExecutionPlan
- Rapier simulation (client)
- MuJoCo validation (server)
- Export bundle generation
| Layer | Technology |
|---|---|
| Frontend | React 18, TypeScript, Vite 7 |
| 3D and Physics (client) | React Three Fiber, Drei, Rapier WASM |
| Visual Task Editor | React Flow (@xyflow/react) |
| State Management | Jotai |
| AI | Gemini via @google/generative-ai |
| Backend | FastAPI, Uvicorn |
| Validation Physics (server) | MuJoCo 3.x |
| Export Pipeline | Jinja2 templates, deterministic generators |
| E2E Testing | Playwright |
npm install --legacy-peer-deps
pip install -r server/requirements.txtCreate a local environment file from .env.example and provide your keys and secrets.
Required values include:
- GEMINI_API_KEY
- JWT_SECRET
- DATABASE_URL
- VITE_GEMINI_API_KEY (optional but recommended for direct frontend Gemini flow)
npm run devFrontend default URL: http://localhost:5173
cd server
python -m uvicorn main:app --reloadBackend default URL: http://localhost:8000
npm run dev
npm run build
npm run preview
npm run test:e2emirai/
|- src/ # Main React + TypeScript app
| |- components/ # Designer, task editor, simulation, community, export UI
| |- utils/ # Planning, kinematics, compiler, AI and export helpers
| |- store/ # Jotai atoms for app, AI, simulation, MuJoCo, tasks
| |- data/ # Community tasks, presets, BOM catalog
| |- types/ # Arm/task/simulation/AI/mujoco contracts
| |- App.tsx
|- server/ # FastAPI backend
| |- main.py
| |- models/
| |- mujoco/
| |- export/
| |- tests/
| |- requirements.txt
|- e2e/ # Playwright tests
|- docs/ # Static docs site assets
|- miraistaticpage/ # Marketing/landing static page
|- mirai-deck.html # Pitch deck
- Event: Transforming Enterprise Through AI
- Track: Track 3 - Robotics and Simulation
- Build phase: May 11-19, 2026
- Submission deadline: May 19, 2026, 8:00 AM PST
This project is licensed under the MIT License. See LICENSE for details.