jev-autopilot
Three.js drone simulator flown by Jev: code computes the situation, Jev picks stick inputs, and safety limits stay in code.
Overview
Drone Sim with Jev Autopilot is a 3D drone simulator built with Three.js in which Jev flies a drone from pad A across a randomly generated city to land on pad B, avoiding buildings. The hellogumbo list reports a trip costs about $0.01.
The split of work is explicit. Code does the maths: altitude, bearing to the pad, speed, obstacles ahead and the tallest building on the path, turned into a short plain-language situation report. Jev makes the call: one request asks six questions, a Choice per stick axis (throttle, yaw, pitch, roll) and two Booleans (commit to landing, cut motors), in about 200 ms. Code then converts each probability distribution into a stick deflection, with hard safety limits kept in code. Users can fly manually with keyboard or gamepad and take control back by touching any stick; seeds make cities replayable, and the landing banner shows tokens and decisions used.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | How should the throttle, yaw, pitch and roll sticks move? | — |
| Yes / no | Commit to landing? | yes · no |
| Yes / no | Cut motors? | yes · no |
How it works
- Code computes flight state and writes a situation report
- Jev answers six questions in one request
- Probabilities become stick deflections
- Safety limits applied in code; repeat until landed
Strengths
- Clear separation of maths, judgment and control
- Manual override at any time
- Cheap per flight per the list
Caveats
- Simulation only
- Requires a TypeSafe key
Getting started
`npm install`, put TYPESAFE_AI_API_KEY in .env, `npm run dev`, open http://localhost:5173 and press J.At a glance
- Type
- App
- Status
- prototype
- Language
- TypeScript
- Runtime
- browser (Three.js) with server
- Has code
- yes
Use cases and integrations
How curated lists describe it
This demo uses Jev from TypeSafe AI to autonomously fly a drone in a random city from point A to point B, avoiding obstacles along the way. A trip costs $0.01.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev control?
Stick inputs on four axes plus whether to commit to landing or cut motors.
Who does the maths?
Code computes altitude, bearing, speed and obstacles; Jev only judges.
Can I take over?
Yes, touching any stick gives control back.
Related projects
Simulated Franka arm that follows plain-English goals by having Jev chain about 30 hardcoded motion primitives, one per step.
Agent skillprototypePythonCamera-only MuJoCo quadrotor where Jev makes 2.5 Hz tactical judgments while code keeps flight control and a safety veto.
ProjectprototypePythonMulti-drone simulation where Jev makes fast reflex decisions per drone and an optional slower LLM may advise but never takes control.
AppprototypeTypeScriptOpen 4B System I driving model that scores behaviours like cruise, stop, yield and brake from three camera frames each decision.
ProjectresearchPythonExperimental robot loop where Jev chooses the next verified action from perception-generated options for a language goal.
ProjectresearchPythonMuJoCo PiPER arm controlled by a local decider-2b model choosing among ten skills via a Jev-style typed-choice interface while targets move.
Agent skillresearchPython