Jev Self-Driving Sim
Browser 2D driving sim where Jev answers lane, speed, hazard and pedestrian-yield questions every 200 ms, with an LLM comparison mode.
Overview
Jev Self-Driving Sim is a top-down autonomous car in the browser driven by Jev. About every 200 ms the car turns its sensor readings into a JSON state and sends four typed questions in one call: lane action (keep, change left, change right), speed action (stop, slow down, hold, speed up), hazard score from 0 (clear) to 3 (collision likely within seconds), and a pedestrian-yield Noul. Code applies confidence-gated routing: low-confidence lane changes are ignored, a strong yield overrides speed, severe hazard forces at least slowing, and stops soften to slowing when nothing is close. A small code reflex for emergency braking and blind-spot aborts can be switched off.
With an OpenRouter key, a compare mode runs two tracks on the same seeded course, one driven by Jev and one by an LLM (DeepSeek V4.1 Flash by default) receiving the same state and questions, with a live table of distance, speed, decisions, latency, tokens and cost per decision, kilometre and hour. Preset courses such as slalom, chicane and gauntlet force steering. There was no licence file at the time of writing.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | lane_action: what should the car do with its lane? | keep_lane · change_left · change_right |
| Choice | speed_action: how should the car adjust speed? | stop · slow_down · hold · speed_up |
| Score | hazard: how likely is a collision (0 clear to 3 imminent)? | 0 · 1 · 2 · 3 |
| Yes / no | pedestrian_yield: must the car stop for a pedestrian? | yes · no |
How it works
- Build JSON state from sensors every ~200 ms
- Ask four questions in one Jev call
- Apply confidence-gated rules in code
- Optional code reflex handles imminent impacts
- Compare with an LLM driver on the same course
Strengths
- Clear example of confidence-gated control
- Side-by-side cost and latency comparison
- Seeded, repeatable courses
Caveats
- Simulation only
- No licence file at time of writing
- Requires a TypeSafe key (and OpenRouter for compare)
At a glance
- Type
- Project
- Status
- prototype
- Language
- JavaScript
- Runtime
- Browser app
- Has code
- yes
Use cases and integrations
How curated lists describe it
A browser-based top-down driving simulator using Jev for lane and speed choices, with an optional chat-model comparison.
logicrw/awesome-jev-projects2D autonomous car simulation in the browser, driven by TypeSafe's Jev decision model.
hellogumbo/awesome-jevA 2D top-down car in the browser that turns its sensors into a JSON state every 200 ms and executes four typed answers. No license file at the time of writing.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
Lane action, speed action, hazard level and whether to yield to a pedestrian.
Can I compare it with an LLM?
Yes. Compare mode runs an LLM on an identical course with the same questions.
Is there a safety fallback?
A code reflex handles imminent impacts and can be disabled.
Related projects
Experimental Conway's Life game where Jev makes each move with one boolean per legal cell, plus a broker that wraps any LLM as Jev.
ProjectprototypeTypeScriptEvaluates Jev as a fixed policy in Gymnasium and Atari games, picking a legal action each step via one Choice question.
ProjectresearchPythonModel Kombat: a 12,000-neuron fly connectome fights Jev in mk.js, with Jev picking one of seven moves about five times a second.
ProjectresearchPythonHarness where Jev plays the first StarCraft shareware mission from structured state, with recorded per-decision action probabilities.
BenchmarkresearchFifteen live arcade games where Jev and LLMs play each other, led by WikiRace in which Jev picks only real links.
ProjectbetaTypeScriptReal-time bot battles where you write a fighter's strategy in plain English and a System One model pilots it several times a second.
ProjectbetaTypeScript