jevbooks

← All projects

jev-drone

RomanSlack/jev-drone

Camera-only MuJoCo drone with Jev judgment model in the loop

Agent Decisions67%Runner-up: Classification & RoutingMany questions, one request

What it is

An autonomous quadrotor flies a five-station obstacle course in MuJoCo using only its onboard camera. Classical CV and control stay in code; a small Jev model decides what each situation means. For robotics researchers exploring judgment models in live control loops.

How it uses Jev

At ~2.5 Hz, on a symbolic scene of five range sectors, blocker height, top-edge visibility, and target position, Jev answers three questions in one call: maneuver (Choice: hold_course/gap_left/gap_right/climb/brake/reacquire), risk (Score), and target_truly_lost (Noul). Code decides when to ask; a 50 Hz reflex layer can override any judgment. Jev is advisory only.

Primitives:choicescorenoul

Technique worth stealing

Include the variable the answer depends on in the state: adding obstruction top-edge height and climb ceiling moved 'climb' from never-chosen to p=0.93.

Try it

./setup.sh, cp .env.example .env with your key, then .venv/bin/python run.py --seconds 65 --seeds 1

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.78
  • Shows a System One patternyes0.81
  • Handles uncertaintyno0.08
  • Measuredno0.17
  • Runnableno0.27
  • Worth recommendingunclear0.51
  • Model replicano0.05
  • Problem scopescore on a 0–2 scale0.12
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.