DriveJev
Open 4B System I driving model that scores behaviours like cruise, stop, yield and brake from three camera frames each decision.
Overview
DriveJev is an open-source System I decision model for autonomous driving. At every decision it looks at three camera frames and a short description of the situation, scores the behaviours the car can carry out at that instant (cruise, stop at the line, hold, start, turn, yield, brake) and returns a probability for each in one forward pass, without writing out reasoning or planning a trajectory. A local executor turns the chosen behaviour into steering and speed.
It drives closed-loop in the JevPilot simulator across town, city and interstate scenes with traffic, pedestrians, lights, stop signs and scripted interactions. The README reports that on held-out tests it drives 28 of 35 interaction episodes and 25 of 28 base episodes cleanly, meets 241 of 242 scripted hazards without contact, and decides in 111 ms on one GPU. Code, the DriveJev-4B weights, the harness and a live demo are released.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which behaviour should the car carry out now? | cruise · stop at the line · hold · start · turn · yield · brake |
How it works
- Capture three camera frames and a situation description
- Model scores each offered behaviour
- Local executor converts the chosen behaviour to steering and speed
- Repeat each decision step in closed loop
Strengths
- Fast decisions (111 ms on one GPU reported)
- Control stays in a deterministic executor
- Open weights, harness and demo
Caveats
- Simulation only; not validated on real vehicles
- Some interaction episodes still fail
Getting started
See the Hugging Face model benmagnifico/DriveJev-4B and the JevPilot harness.At a glance
- Type
- Project
- Status
- research
- Language
- Python
- Runtime
- PyTorch + JevPilot simulator
- Has code
- yes
Use cases and integrations
How curated lists describe it
An open System One decision model for autonomous driving.
MrJev/awesome-jev
Frequently asked questions
What does DriveJev decide?
Which driving behaviour (cruise, stop, hold, start, turn, yield, brake) to perform at each step.
Who controls steering?
A local executor turns the chosen behaviour into steering and speed.
Is it road-ready?
No, it is evaluated in the JevPilot simulator.
Related projects
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- EmbodiedJev★ 302
Browser MuJoCo Franka Panda workbench comparing Jev, Claude, OpenAI-compatible and local models on pick, stack and obstacle tasks.
ServiceresearchPython Simulated Franka arm that follows plain-English goals by having Jev chain about 30 hardcoded motion primitives, one per step.
Agent skillprototypePythonCamera-only MuJoCo quadrotor where Jev makes 2.5 Hz tactical judgments while code keeps flight control and a safety veto.
ProjectprototypePythonMulti-drone simulation where Jev makes fast reflex decisions per drone and an optional slower LLM may advise but never takes control.
AppprototypeTypeScriptExperimental robot loop where Jev chooses the next verified action from perception-generated options for a language goal.
ProjectresearchPython