Back
roboticsresearchOpen source

jev-drone

Use Jev for tactical judgments in a camera-driven drone simulation.

What Jev decides

Which maneuver fits, how risky the scene is, and whether the target is lost.

jev-drone flies a MuJoCo obstacle course with camera-derived scene facts, Jev maneuver advice, and code-owned flight safety.

CHOICESCORENOUL
Jev role
Advise the drone on tactical maneuver and scene risk.
Input
Camera-derived range sectors, obstacle features, and target status.
Output
Choice, Score, and Noul judgments consumed by the guidance code.
Decision frequency
Around 2.5 Hz when the scene warrants a new judgment.
Fallback
Local guidance and safety reflexes retain control and can veto Jev advice.
Author reported · Source reviewed

How it works

Input
Camera-derived range sectors, obstacle features, and target status.
Jev
CHOICE + SCORE + NOUL
Output
Choice, Score, and Noul judgments consumed by the guidance code.

Overview

Computer vision turns depth and segmentation images into a compact scene. Jev chooses a maneuver, scores risk, and judges whether the target is truly lost. A faster guidance and reflex layer controls flight, can veto unsafe proposals, and reuses judgments when the scene is unchanged. The README reports a full-course flight and a local baseline comparison; those are author-reported simulation results, not evidence of a physical drone test.

Language
Python
License
MIT
Author
RomanSlack
Sources checked Sep 26, 2026

Sources

Related projects