Mobileye Patents a System That Draws a Safe Inner Path Inside Every Lane
A lane on a map tells a self-driving car where it's allowed to go. But Mobileye's new patent describes something more precise: a narrower "sleeve" drawn inside that lane, calculated from real vehicle trips, that tells the car exactly where it should go.
What Mobileye's virtual lane sleeve actually does for drivers
Every time a car drives down a road, it leaves behind a record of its exact path. A self-driving system that ignores all that data is essentially navigating blind.
Mobileye's patent describes a process that collects real-world driving data from vehicles that have already traveled a road, then uses that data to reconstruct the lane's edges. From there, it creates what the patent calls a virtual sleeve -- a tighter corridor drawn inside the lane boundaries. Think of it like a painted bike lane inside a regular traffic lane: you're technically allowed anywhere in the big lane, but the sleeve shows where you actually want to be.
That sleeve becomes a drivable path stored in a shared map, which gets sent out to vehicles equipped with Mobileye's navigation system. Instead of each car figuring out the best path from scratch, the map does that work once and shares it with everyone. It's a way of turning collective driving experience into a reliable, reusable guide.
… determine a left side boundary and a right side boundary of a virtual sleeve associated with the lane of travel, wherein an area associated with the virtual sleeve over a section of the road segment differs from an area associated with the lane of travel over the section of the road segment; …
Translation: The system draws a narrow safety zone inside the actual lane bounds.
How the system builds and stores a drivable path from real trips
The system starts by receiving drive information from a vehicle that has already traveled a given road segment. That information includes what the patent calls road topography indicators -- data about the physical shape and features of the road, like curves, elevation changes, or lane markings.
Using those indicators, the system reconstructs a representation of the lane's left and right boundaries -- essentially, a digital picture of where the lane begins and ends. Then it computes a virtual sleeve: a secondary boundary pair that sits inside the lane boundaries but does not necessarily match them. The area enclosed by the sleeve is intentionally different from the full lane area, giving the system room to define a path that is safer or more comfortable than simply splitting the lane down the middle.
From the sleeve's left and right edges, the system generates a vehicle drivable path -- a specific line or corridor that a self-driving car should follow. That path is stored in a shared map.
The map is then distributed to host vehicle navigation systems, meaning multiple cars can benefit from the path without each one having to independently calculate it. The key components of the pipeline are:
- Data collection from prior vehicle trips
- Lane boundary reconstruction from topography data
- Virtual sleeve generation inside those boundaries
- Drivable path derivation and map storage
- Map distribution to other vehicles
… receive drive information from a vehicle that previously traversed the road segment.
Translation: It uses historical driving data from cars that drove that road before.
What this means for self-driving cars on tricky roads
For self-driving cars, the difference between a lane boundary and a genuinely safe driving path is significant. A lane might be fifteen feet wide, but a car that hugs one edge or weaves through the middle is more likely to clip a curb, brush a guardrail, or unsettle other drivers. A pre-computed path that accounts for actual road shape reduces that risk.
The crowd-sourced map model is also worth noting. Rather than relying solely on expensive sensor sweeps or satellite imagery, this approach lets ordinary vehicle trips continuously refine the map data. That could make the system more adaptable to real-world conditions -- potholes, shifting lane markers, construction zones -- than a static map updated infrequently. For anyone riding in a vehicle running on Mobileye hardware, this is the kind of background infrastructure that shapes whether the car feels confident or jittery on unfamiliar roads.
This is the 151st Intel filing in our Intel coverage since May, adding to work like faster chip math and handling shifting data sizes.
The core trade-off in this design is that the system relies on historical driving data to build paths for future vehicles. That works well on roads that get regular traffic, but it creates a cold-start problem on rarely traveled routes: if no vehicle has driven a road recently, or if the available trips were made by drivers who took unusual lines, the generated sleeve could be misleading rather than helpful.
The sleeve concept also introduces a calibration question the patent does not fully resolve. How much narrower than the lane should the sleeve be? On a wide highway, a tight sleeve probably makes sense. On a narrow country road, shrinking the drivable path further could leave almost no margin for error. The answer likely depends on road type, speed, and surrounding conditions, and getting that calibration wrong in either direction has real consequences.
That said, the map-distribution model is the genuinely practical idea here. Pre-computing a path once and sharing it across a fleet is far more efficient than asking each car to solve the same geometry problem independently. Mobileye's long bet on crowd-sourced road intelligence shows up here in a concrete, testable form, and this patent gives it a cleaner engineering shape than most prior approaches in this space.
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
53 drawing sheets from US 2026/0276397 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →
Be the first to weigh in