Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

Multi-UAV-JEV

Camera-observed cooperative drone pursuit in a MuJoCo city where each UAV's tactical move is a /v1/systemone decision.

ServiceresearchApache-2.0
View on GitHub ↗

Overview

Multi-UAV JEV simulates cooperative drone pursuit in a MuJoCo city with 1, 2 or 4 pursuers. Each aircraft has its own physics, four motors, camera perception, controller and a decision worker that calls a /v1/systemone endpoint; per the list summary the default is a local NeoHorse-Jev-4B server on a CUDA GPU. Roles are complementary: continuous tracking, forward blockade, predictive interception, and communication support with rescue.

Pursuers share camera-derived observations with confidence and timestamps, and stale observations expire into search. Role rules live in an editable Markdown file that reloads while running. A live cockpit shows RGB, depth, candidate probabilities, action counts and predicted paths, and the system records 20 FPS replays and archives successful and failed decision logs. It is a research and demo platform for tactical decisions in robotics simulation.

What the decision model decides

TypeQuestionAnswers
ChoiceNext tactical action for this pursuer given its role and observations(inferred)—

How it works

  1. Each UAV perceives the target through its camera.
  2. Observations are fused and shared across the fleet.
  3. Each aircraft's worker asks the decision model for its next action.
  4. Controllers execute; replays and decision records are archived.

Strengths

  • Can run against a local decision server, avoiding API calls.
  • Candidate probabilities are visible in the HUD.
  • Editable role rules hot-reload during a run.

Caveats

  • Simulation only; local default needs a CUDA GPU.
  • Licence not stated in the README excerpt (list says Apache-2.0).

At a glance

Type
Service
Status
research
Language
Python
Runtime
MuJoCo simulator
Has code
yes
License
Apache-2.0

Use cases and integrations

How curated lists describe it

  • Camera-observed UAV pursuit in MuJoCo with Jev decisions, RGB/depth HUD, replay and episode collection.

    hellogumbo/awesome-jev

  • Camera-observed drone pursuit in MuJoCo where each decision is a `/v1/systemone` call, by default to a local NeoHorse-Jev-4B server (BF16 CUDA GPU), with an RGB/depth HUD, replays and episode collection. Apache-2.0.

    MrJev/awesome-jev

Frequently asked questions

What does the decision model choose?

Each pursuer's next tactical action given its role and the shared observations.

Does it need TypeSafe?

Not by default; per the list summary it calls a local NeoHorse-Jev-4B server over /v1/systemone.

Can I change drone behaviour?

Yes, edit sim/role.md; valid edits reload while running.

Related projects

  • jev-askable-arm

    Simulated Franka arm that follows plain-English goals by having Jev chain about 30 hardcoded motion primitives, one per step.

    Agent skillprototypePython
  • jev-drone

    Camera-only MuJoCo quadrotor where Jev makes 2.5 Hz tactical judgments while code keeps flight control and a safety veto.

    ProjectprototypePython
  • 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.

    AppprototypeTypeScript
  • DriveJev

    Open 4B System I driving model that scores behaviours like cruise, stop, yield and brake from three camera frames each decision.

    ProjectresearchPython
  • JEV Control Your Roboarm

    Experimental robot loop where Jev chooses the next verified action from perception-generated options for a language goal.

    ProjectresearchPython
  • Jev Robot

    MuJoCo 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