jev-pilot-reflex
Three.js driving simulator with a fast reflex layer, a TTC safety brake and an optional Jev API strategy layer.
Overview
JevPilot Reflex is a Three.js 3D driving simulation lab, documented mainly in Chinese, built around a System 1 / System 2 split. It argues that vision-language models take 300-800 ms per decision, during which a car at highway speed travels 10-26 metres. The simulator uses a detailed vehicle model with Ackermann steering, procedural city streets and a highway.
The System 1 layer is a built-in heuristic reflex solver with multi-candidate trajectory evaluation running at 1.5 ms per step, fully offline without an API key. A safety brake based on time-to-collision forces full braking below a critical threshold regardless of higher-level planning. A slower System 2 interface, compatible with a remote Jev API or a VLM, offers strategic advice asynchronously without blocking the 60 FPS control loop. The README describes stages combining vision-language-action, world-model rollouts and language intent with this safety shield. Note that the fast 'Jev reflex' is a local heuristic rather than the hosted Jev model.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which candidate trajectory should the car follow?(inferred) | — |
| Other | Strategic advice from the slow layer (remote Jev API or VLM)(inferred) | — |
How it works
- Simulate the scene and vehicle at 60 FPS
- The local reflex solver scores candidate trajectories each step
- The TTC safety brake overrides when collision time drops below threshold
- An optional remote System 2 gives asynchronous strategy advice
Strengths
- Runs fully offline in the browser
- Safety brake has override authority over planners
- Shows why slow models should not sit in tight control loops
Caveats
- The fast reflex is a heuristic, not the Jev model
- Simulation only
- Chinese-language documentation
At a glance
- Type
- Project
- Status
- prototype
- Language
- JavaScript
- Runtime
- Browser (Three.js)
- Has code
- yes
Use cases and integrations
How curated lists describe it
Autonomous vehicle simulation: Three.js autonomous driving reflex and AI safety brake simulator using Jev System 1/2 dual-brain architecture for fast emergency intervention.
yibie/awesome-jev
Frequently asked questions
Does it call Jev for every frame?
No. The 60 Hz reflex is a local heuristic solver; Jev or a VLM is an optional slow advisory layer.
What is the safety brake?
A time-to-collision check that forces full braking below a threshold, overriding planners.
Do I need an API key?
No for the core simulation; only the optional System 2 layer uses a remote API.
Related projects
Three.js box steered by plain-language instructions: an LLM plans steps, Jev decides move, turn and jump in real time.
ProjectprototypeTypeScriptThree.js driving simulator where Jev picks among locally sampled steering-and-speed paths while code handles geometry and safety.
ProjectprototypeTypeScriptTraffic signal controller where Jev picks each intersection's next phase from a SUMO snapshot, with local fallbacks.
ProjectresearchPythonFixed-wing landing simulator and benchmark where Jev maps four Choice decisions to aircraft controls against humans, baselines and LLMs.
DemoprototypePythonSimulation where Jev picks which of two conflicting signal phases gets green at a real Hong Kong junction from live open sensor data.
ServiceprototypePythonSimulated Franka arm that follows plain-English goals by having Jev chain about 30 hardcoded motion primitives, one per step.
Agent skillprototypePython