Skip to content

Little Airways

Toy archipelago air-traffic sim where each plane asks Jev every 1.6 s whether to divert, declare an emergency, give way or land.

DemoprototypeMIT

Overview

Little Airways is a show-and-tell capability study for Jev: a golden-hour island world where airliners and small props fly between airports and every in-flight judgment is made live by Jev in about 150 ms. Each aircraft knows only about itself and nearby traffic. Roughly every 1.6 seconds it asks four questions in one request: route (continue, divert to an airport or hold), whether to broadcast an emergency, a response to nearby traffic (maintain, give way, relay a mayday, hold or divert) and clearance (approach or go around). The tower asks Jev to sequence landings when planes queue for one runway. Visuals and simulation code were made with other AI models.

The README's lessons are the most useful part. With terse options, Jev split 0.46 give-way to 0.41 maintain for a conflicting emergency, but operationally worded criteria moved it to 0.70 versus 0.05, so the demo switches to richer wording when a mayday is near. Sending raw fuel fractions led Jev to keep cruising; sending estimates such as fuel minutes remaining and minutes to each airport let it make the comparison itself. A monitor shows every raw request and response with latency, tokens and cost.

What the decision model decides

TypeQuestionAnswers
ChoiceShould this aircraft continue, divert to an airport, or hold?continue · divert · hold
Yes / noShould this aircraft declare an emergency?—
ChoiceHow should this aircraft respond to nearby traffic?maintain · give way · relay the mayday · hold · divert
ChoiceWhich aircraft lands first?—

How it works

  1. Each plane builds a local state with fuel and timing estimates
  2. Every ~1.6 s it sends four questions to Jev
  3. Simulation applies route, broadcast, response and clearance answers
  4. Tower asks Jev to order landings
  5. Monitor shows raw requests, latency and cost

Strengths

  • Clear lessons on question wording and state design
  • Raw API traffic visible
  • No scripts or behaviour trees

Caveats

  • Toy simulation, not real ATC
  • Many API calls over time
  • Visual and sim code generated by other models

At a glance

Type
Demo
Status
prototype
Language
TypeScript
Runtime
browser simulation
Has code
yes
License
MIT

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does Jev decide?

Each plane's route, emergency broadcast, response to traffic, clearance, and the tower's landing order.

What was learned about wording?

Operationally worded options moved a give-way decision from a coin flip to 0.70 versus 0.05.

Why send estimates instead of raw data?

With only a fuel fraction Jev kept cruising; with minutes remaining it could compare and divert.

Related projects

  • Korean RL tycoon game training a company decision model that answers situations with choice, probabilities and confidence.

    ProjectprototypePython
  • Interactive Jev experiments, from a support-message decision lab to 3D self-driving demos with steer, speed and overtake decisions.

    Appprototype
  • Real-time bot battles where you write a fighter's strategy in plain English and a System One model pilots it several times a second.

    ProjectbetaTypeScript
  • Bot that plays Red Alert 2 web skirmishes, with code handling mechanics and a decision model making stance, threat and timing calls.

    ServicebetaPython
  • Multiplayer 3D science-fantasy version of Houston's River Oaks District with Jev-powered NPC companion, chauffeur and smart drive.

    ProjectbetaTypeScript
  • Browser 3D planetary rover sandbox where a TypeSafe controller chooses the next action from complete candidates, versus a baseline.

    ProjectbetaTypeScript