Jev HighwayEnv: 60 Seconds Without a Crash

Description and use

Demonstrated Jev AI model in HighwayEnv: 60‑second autonomous drive without collisions, optimizing speed ranges, stop rules, and latency.

Problem

Achieving a 60‑second autonomous drive on a three‑lane highway without collisions using the Jev AI model, while managing latency, speed limits, and stopping behavior.

Outcome

The car achieved a 60‑second run covering 1,203.72 m at an average speed of 72.3 km/h with no collisions or stops, using Jev’s 311 ms median response latency.

Workflow

• 1. Define state and questions for Jev (collision risk for each action).
• 2. Send state and questions to Jev via TypeSafe SDK.
• 3. Receive risk predictions for IDLE, SLOWER, LANE_LEFT.
• 4. In code, compare risks and apply action if risk ≤0.3.
• 5. Update target speed and repeat each simulation step.
• 6. Monitor for STALLED condition (speed <1 m/s for 10 s).
• 7. Separate physics (30 Hz) from rendering (≤10 fps) to reduce latency.

Evidence

The car completed the run at an average speed of 20.08 m/s (about 72.3 km/h), without stopping to run down the clock.

Source

Response latency (p50 / p95): 311 ms / 482 ms

Source

I expanded the target speeds to 0–30 m/s in 5 m/s increments, allowing it to slow down in steps.

Source