Digital ATC

Flight simulation for AI development

Build LLMs and algorithms without fighting with overly complex simulators

---

What This Is

Digital ATC is an open-source flight simulation platform built specifically for developing AI systems, algorithms, and LLMs. Think of it as the Goldilocks zone: realistic enough to be useful, simple enough to iterate fast.

The Problem We're Solving

Here's the thing: existing flight simulators are way too realistic for what most of us actually need. When you're building:

• LLMs that understand ATC instructions
• Coordination algorithms for multi-aircraft scenarios
• Navigation systems for autonomous flight
• Decision-making algorithms for conflict resolution

...you don't need perfect aerodynamics. You need:

• Fast iteration - test ideas in minutes, not hours
• Simple setup - runs in a browser, no WSL, no Linux headaches, no RAM-hungry installations
• Configurable parameters - set max speed, vertical speed limits, turn rates without diving into physics textbooks
• Realistic enough - good enough for your use case, simple enough to actually use

Realism is for control algorithms. We're building for AI and decision-making.

What We Built

A lightweight, web-based flight simulator that gets out of your way:

• 🎯 LLM-Powered Digital Pilots (Our main focus) - Parse ATC instructions, generate proper readbacks with ICAO/NATO phonetics, execute commands intelligently
• 🧠 AI Algorithm Development - Perfect for training coordination, navigation, and decision-making systems without physics getting in the way
• ⚡ Simplified 3D Simulation - Point-mass physics over real Mapbox terrain (San Francisco Bay Area)
• 🔧 Configurable Flight Envelopes - Set max speed, vertical speed limits, turn rates with simple parameters
• 📋 Scenario-Based Training - Pre-built scenarios for rapid testing
• 🌐 Zero-Install Development - Browser-based, works everywhere, no heavy dependencies

Use Cases

Primary: LLM Development

• Train language models to understand ATC instructions
• Generate proper aviation readbacks with phonetics
• Build conversational AI for air traffic control

AI & Algorithm Development

• Coordination algorithms for multi-aircraft scenarios
• Navigation algorithms for autonomous flight planning
• Decision-making systems for conflict resolution
• ATC system training and testing

Rapid Prototyping

• Test ideas quickly without complex simulator setup
• Iterate on algorithms with configurable flight parameters
• Validate concepts before moving to full-fidelity simulators

Quick Start

# Install dependencies
npm install

# Copy the sample environment file and fill in tokens
cp .env.example .env

# Run it
npm run dev

That's it. No Docker, no WSL, no 50GB downloads. Just works.

Features

✅ LLM Integration Ready - Structured I/O schemas and prompt templates for digital pilot development
✅ Configurable Flight Envelopes - Set aircraft parameters programmatically
✅ 3D Terrain Visualization - Mapbox terrain with real-time aircraft tracking
✅ Point-Mass Physics - Simplified dynamics perfect for high-level algorithm work
✅ Manual & Automated Controls - Direct control or target-based automation
✅ Scenario Playback - Pre-built scenarios for testing

Tech Stack

• Frontend: Vue 3 + Vite (runs in browser, zero installation)
• 3D Rendering: Mapbox GL JS + Three.js
• Physics: Custom point-mass engine (configurable, not over-engineered)
• AI Ready: JSON schemas, prompt templates, lexicon for LLM integration

Environment & LLM Setup

• VITE_MAPBOX_TOKEN is required for terrain rendering. Sign up at Mapbox for a free developer token.
• VITE_OPENAI_API_KEY enables the digital pilot. Copy your key from the OpenAI dashboard and paste it into .env.
• VITE_OPENAI_MODEL defaults to gpt-4o-mini, a fast/affordable model (~$0.15 per 1M input tokens, ~$0.60 per 1M output tokens). The default demo flow stays under a cent.
• If either token is missing the app will warn you on startup; Mapbox terrain and LLM automation fall back gracefully.

Default Demo Scenario

• Choose "Default Demo" in the Scenario selector to run a scripted KOAK shoreline flight.
• The timeline injects a realistic mix of ATC calls: climb & heading assignments, TFR avoidance, speed reduction for traffic, and a handoff.
• Every ATC transmission is piped through the digital pilot (src/llm/pilotAgent.js); watch the Transcript tab for live readbacks and the State tab for updated intent/safety flags.
• Scenario data lives in scenarios/Default_KOAK_demo.json and is executed by the runner in src/sim/scenarioRunner.js—tweak or extend events without touching UI wiring.

Current Status

What's Working:

• Core flight simulation with manual controls and automated target following
• LLM-ready architecture with schemas and prompts
• Real-time 3D visualization over Mapbox terrain
• Configurable flight parameters

What's Next:

• LLM integration for ATC parsing and readback generation
• Conflict detection and safety flagging
• Enhanced scenario system

---

MIT License