What it is
A small, inspectable robotics laboratory where a Franka Panda in MuJoCo performs pick & place, pushing, stacking, peg insertion, and gate pick & place using structured simulator state, not images.
How it uses Jev
JEV receives structured simulator state (TCP and object poses, velocity, jaw width, contacts, geometry, recent actions) and makes two-stage decisions: first a Choice over task-specific intentions, then Choice for X/Y/Z directions and gripper command. Malformed responses execute no action; success is checked independently by physics.
Primitives:choice
Technique worth stealing
Two-stage JEV control: intent selection followed by Cartesian motion and gripper commands, with independent physical success checks.
Try it
git clone https://github.com/lykycy123/RoboJEV.git; conda env create -f environment.yml; conda activate jev-vla-sim; python -m pip install -e '.[test,video]'; python scripts/fetch_panda.py; robojev --task pick_place --policy rule --seed 1000
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-centricyes0.96
- Shows a System One patternyes0.69
- Handles uncertaintyno0.05
- Measuredno0.06
- Runnableno0.37
- Worth recommendingyes0.63
- Model replicaunclear0.49
- Problem scopescore on a 0–2 scale0.44
- About Jevyes0.96
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.