Jev with Foresight: Safe Language-Guided Autonomous Driving via a Language World Model
Jev with Foresight: Safe Language-Guided Autonomous Driving via a Language World Model
This research presents Jev-AD, which integrates a typed decision model with a language world model to predict vehicle trajectories over two seconds for safer autonomous driving. In simulated highway merging scenarios involving 500 car episodes, the system reduced the collision rate from 26.0% to 2.4% while maintaining a decision latency of 0.25 seconds per action and achieving passenger intent in 93% of free-form command trials. When evaluated on 500 replayed real-world NGSIM episodes, the collision rate decreased from 48.2% to 2.8%, and the world model successfully foresaw 93% of dangerous actions.
The paper introduces Jev-AD, a system combining a typed decision model with a language world model to predict vehicle movements over two seconds for safer autonomous driving decisions; reported improvements in collision rates are based on simulated highway merging and replayed real-world NGSIM episodes.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
decision time per action
0.25 seconds
View original evidence
or every candidate action how nearby vehicles move over the next two seconds. Code turns these predictions into foreseen consequences, which Jev compares instead of inferring them from coordinates. In simulated highway merging (500 car-episodes), this foresight lowers the same decision model's collision rate from 26.0% to 2.4% at 0.25 seconds per decision, and Jev-AD achieves the passenger's intent in 93% of free-for
Open source S3
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
arXiv:2611.05448v1 Announce Type: new Abstract: Large language models understand passengers' driving commands, but as driving agents they reason in generated text for seconds and must infer each action's consequences implicitly from the scene. We present Jev-AD, which pairs Jev, a typed decision model that answers predefined questions with probabilities, not generated text, with a language world model that predicts f
Open source S2
or every candidate action how nearby vehicles move over the next two seconds. Code turns these predictions into foreseen consequences, which Jev compares instead of inferring them from coordinates. In simulated highway merging (500 car-episodes), this foresight lowers the same decision model's collision rate from 26.0% to 2.4% at 0.25 seconds per decision, and Jev-AD achieves the passenger's intent in 93% of free-for
Open source S3
m command trials. In 500 replayed real-world NGSIM episodes, foresight lowers the collision rate from 48.2% to 2.8%, and the world model, learned from this traffic, foresees 93% of dangerous actions. Project page: https://ys-qu.github.io/jev-ad-website.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org