jev-reflex-autonomy-lab
Multi-drone simulation where Jev makes fast reflex decisions per drone and an optional slower LLM may advise but never takes control.
Overview
Jev Reflex Autonomy Lab is an interactive multi-drone simulation exploring what happens when fast, typed System 1 reflexes can ask a slower System 2 model for advice without giving up control. One to fifteen drones each navigate independently with Jev as the reflex layer, avoiding moving asteroids, debris, unidentified objects and each other on the way to a shared destination.
When confidence falls below a configured threshold, an optional System 2 planner (GLM 5.3 via OpenRouter by default) gives one-use strategic guidance asynchronously; Jev keeps steering while it waits, and the planner never flies a drone directly. The dashboard shows per-drone decisions, confidence, latency, strategy revisions and collisions, with charts that mark when System 2 guidance was actually used. Scenarios are seeded for repeatability, missions export to JSON, and the app starts with mocks so it runs without credentials.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which manoeuvre should this drone take now given sensed contacts?(inferred) | — |
How it works
- Each tick, sense contacts and project actions per drone
- Jev picks an action for every active drone
- If confidence is low, request advice from System 2 asynchronously
- Jev uses the advice once when it arrives
- Log telemetry and collisions
Strengths
- Direct comparison of Jev alone versus with advice
- Seeded scenarios and JSON export
- Runs with mocks without keys
Caveats
- Simulation only
- System 2 needs an OpenRouter key
- Results depend on scenario design
Getting started
npm install && npm run dev (Node 22.13+), then copy .dev.vars.example to .dev.vars for live providers.At a glance
- Type
- App
- Status
- prototype
- Language
- TypeScript
- Runtime
- React web app
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Drone autonomy: a multi-drone lab where Jev supplies the reflex decisions, with an optional slower strategy layer guiding them.
yibie/awesome-jevMulti-drone autonomy lab demonstrating TypeSafe Jev reflex decisions with optional System 2 strategy guidance.
logicrw/awesome-jev-projectsMulti-drone autonomy lab demonstrating TypeSafe Jev reflex decisions with optional System 2 strategy guidance.
hellogumbo/awesome-jevMulti-drone simulation where typed reflexes fly the fleet and an optional slower planner may advise but never takes control.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
Each drone's next action from its sensed situation.
Does the LLM fly the drones?
No. It only gives one-use advice when Jev's confidence is low.
Can I run it without keys?
Yes. It starts with development mocks.
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.
ProjectprototypePythonOpen 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 skillresearchPythonMuJoCo pilot where Jev, Claude Haiku and reactive rules choose pick-and-place actions; Jev matched the rules exactly.
ProjectresearchPython