User repository on GitHub
Fine-grained robot control with Jev, physics previews, and configurable LIBERO tasks.
- Forks
- 2
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
Jev directory
CommunityJev Cookbooks
Jev × LIBERO
A fine-grained robot control study loop for LIBERO tasks: Jev layers an intent, a contact or motion family, and one of 27 control inputs, while reversible MuJoCo previews evaluate each candidate before execution.
- Stars
- 83
- Language
- Python
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- Dimweaker
- Last verified
- Sep 2026
About this project
Explore robot control with Jev, local physics previews, and configurable tasks. Open the interactive Decision Lab ↗. Demos · Quick start · How it works · Results · Task configs · 简体中文. Close the microwave Close the top drawer. 14 decisions · 111 environment steps MP4 · Full record 20 decisions · 155 environment steps MP4 · Full record Grasp and lower alphabet soup into the basket 40 decisions · 314 environment steps MP4 · Full record.
Quoted from the project's README. Read the full README.
Works forJev for Realtime control
Used inJev for Robotics & Hardware
Highlights
- 27 inputs cover Cartesian translations, wrist rotations, gripper open/close, and hold.
- Layered decisions: Jev selects an intent, a contact or motion family, and an input, each choice informing the next.
- Preview branches run for up to 8 environment steps or 0.4 simulation seconds before a control is executed.
- Runs against the official API with jev-latest or OpenRouter's typesafe/jev-1.13, the CLI default.
- Recorded episodes: microwave in 14 decisions and 111 steps, top drawer in 20 and 155, alphabet soup in 40 and 314.
Quickstart
bash
git clone https://github.com/Dimweaker/jev-libero.git
cd jev-libero
pip install -e .
export TYPESAFE_API_KEY="YOUR_API_KEY"
jev-libero run --provider typesafe --task top_drawer --seed 1 \
--out runs/drawer-s1 --max-decisions 100 --budget-usd 0.10Watch out
MIT-licensed; requires Python 3.10 or 3.11 plus a LIBERO, robosuite, and MuJoCo environment and a TypeSafe or OpenRouter key, and episodes make paid API calls guarded by --budget-usd.
Questions about this entry
Short answers drawn from the linked project or page.
It picks from a menu, but the menu is rebuilt from the simulator every step. Jev selects an intent, then a contact or motion family, then one of 27 inputs covering Cartesian translations, wrist rotations, gripper open/close and hold, with each choice informing the next. It never emits continuous control directly; the controller executes one selected input and observes the result.
That is the point of it. The engine branches the simulator reversibly and previews each candidate input for up to 8 environment steps or 0.4 simulation seconds, with MuJoCo supplying dynamics and FCL measuring collision-shape distances. When a useful move needs repositioning first, two-step previews look for a route to the desired effect.
Yes. Every semantically routed decision is a paid API call through the official TypeSafe API or OpenRouter, and the CLI bounds a run with --max-decisions and a client-side --budget-usd guard. Recorded episodes cost about $0.0012 for the microwave, $0.0014 for the drawer and roughly $0.003 for alphabet soup, though TypeSafe costs are estimated from token usage rather than reported by the provider.
One seed each, and the README is upfront about it. The table lists a single recorded episode per bundled task, all using saved initial-state index 0: microwave passing in 14 decisions, top drawer in 20, alphabet soup in 40. No rate across seeds is reported. A replay command re-applies saved controls and checks the outcome without any API calls.
Use Python 3.10 or 3.11: `git clone`, `pip install -e .`, then connect a LIBERO environment with `pip install python-fcl scipy`, `export LIBERO_ROOT=/path/to/LIBERO` and `export MUJOCO_GL=egl`. Running episodes needs a TypeSafe or OpenRouter API key.
The engine previews each candidate input for up to 8 environment steps or 0.4 simulation seconds: MuJoCo supplies the dynamics, FCL measures collision-shape distances, and the task configuration decides which effects qualify. Two-step previews look for a route when repositioning is needed first.
MIT. It is built on LIBERO, robosuite, MuJoCo, python-fcl and TypeSafe Jev, and the decision interface takes inspiration from TypeSafe Mario.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Pick Verify: a single decision at the warehouse pick station catches confusable picks.
Show HN: Verifying robot picks with a decision modelRead the thread on Hacker NewsX: A dual-arm robot with Jev in the middle layer
YouTube: Dan Makes Things drives an ESP32 robot with Jev decisions.
More Jev resources like this
A MuJoCo and Franka Panda workbench where Jev chooses the next manipulation step for pick-and-place, stacking, and obstacle-carry tasks, with rule baselines and side-by-side model comparison.
Jev Cookbooks#community#python#robotics
Community
FBddcz/embodied-jevPython
FBddcz/embodied-jevPythonA camera-only MuJoCo quadrotor that uses Jev for tactical maneuvers at 2.5 Hz while classical perception and a 50 Hz reflex layer keep safety outside Jev, clearing a five-station course.
Jev Cookbooks#community#python#robotics
Community
RomanSlack/jev-dronePython
RomanSlack/jev-dronePythonA ManiSkill demo where plain-English goals drive a simulated Franka Panda: Jev picks one primitive per step from a catalog of about 30 hardcoded skills, and Python executes the move. No training or policy.
Jev Cookbooks#community#python#robotics
Community
TarunTomar122/jev-askable-armPython
TarunTomar122/jev-askable-armPythonAn experimental controller that has Jev choose NES inputs to play Super Mario Bros. from structured emulator telemetry instead of screenshots, logging every decision with probabilities and latency.
Jev Cookbooks#community#python#games
Community
fhshaik/typesafe-marioPython
fhshaik/typesafe-marioPythonTwo-stage Jev control of a Franka Panda in MuJoCo: Jev selects a task intent, then X/Y/Z directions and a gripper command executed against real contacts, with 43 of 50 Jev episodes succeeding versus 48 of 50 for a rule baseline.
Jev Cookbooks#community#python#robotics
Community
lykycy123/RoboJEVPython
lykycy123/RoboJEVPythonAn SO-101 workbench where an evaluation model or rules baseline picks bounded joint steps from typed candidate actions: decision runs are spend-capped via the Vercel AI Gateway and drive a real LeRobot arm or mock hardware.
Jev Cookbooks#community#typescript#robotics
Community
grmkris/robo-harnessTypeScript
grmkris/robo-harnessTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.