Nvidia Patents an AI That Spots Distracted Drivers by Tracking Their Eye Movements Over Time
A single glance away from the road is easy to miss. Nvidia's new patent describes an AI that watches not just where your eyes are pointing right now, but how your gaze has been moving over the past several seconds, and flags when something looks wrong.
How Nvidia's eye-tracking distraction detection works
Imagine a co-pilot who watches your eyes the whole time you drive. If you spend too long staring at your phone, zone out into a fixed stare, or keep scanning in an unusual pattern, that co-pilot would notice. That's essentially what Nvidia is building here.
The system uses a camera pointed at the driver, captures a series of images over a short window of time, and extracts gaze features, information about where and how your eyes are moving. It then feeds that sequence into an AI model trained to recognize what distracted eye behavior looks like, and flags a warning if your pattern seems off.
The key word is "sequence." Most basic driver-monitoring systems just check a single moment: are your eyes open? Are they facing forward? Nvidia's approach looks at a pattern over time, which is much harder to fool and more likely to catch real distraction before an accident happens.
How the AI reads a sequence of gaze data
The patent describes a pipeline with three main stages.
- Image capture: A camera records two or more frames of the driver's face during a defined time window. In practice this would be a continuous video feed.
- Gaze feature extraction: The system processes those frames to calculate gaze features (the direction each eye is pointing, how quickly the gaze is moving, where it fixates, and so on) and assembles them into a time-ordered sequence.
- AI classification: That sequence is fed into a trained AI model, which has learned to distinguish normal driving attention from abnormal gaze patterns. The model outputs an indication of whether the pattern is abnormal, and the system uses that output to decide whether the driver is distracted.
The emphasis on a time window is what separates this from simpler eye-tracking. A driver might glance down briefly for a perfectly safe reason; only a sustained or repeated pattern would trigger a distraction flag. The AI is trained to understand the difference, though the patent doesn't specify exactly which model architecture is used.
What this means for driver-safety tech in cars
Driver monitoring systems are already required in new vehicles sold in Europe, and automakers worldwide are racing to make them more accurate. A system that tracks gaze patterns over time rather than point-in-time snapshots is meaningfully harder to fool and more sensitive to the kinds of distraction that cause real crashes, like drowsy staring or prolonged phone use.
Nvidia is already a major supplier of the computing hardware inside vehicles through its DRIVE platform. A software capability like this would fit naturally into that ecosystem, giving automakers an AI-powered safety layer they can bundle with Nvidia's existing in-car chips rather than sourcing from a separate vendor.
This is a focused, practical patent in a space where Nvidia has real business momentum. It's not a moonshot concept, driver monitoring is a live commercial requirement right now, and a time-aware gaze-pattern approach is a genuine step up from the frame-by-frame methods already on the market. Worth watching as a sign that Nvidia is pushing its DRIVE platform up the safety-feature stack.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
19 drawing sheets from US 2026/0229046 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →
Editorial commentary on a publicly published patent application. Not legal advice.