Intel's Self-Driving Division Files Patent to Fix Gaps Between Digital Map Sections
Self-driving maps are built in pieces, and the edges of those pieces do not always line up. Mobileye describes a way to pull them together like stretching a sheet of rubber, with built-in resistance so the rest of the map does not tear out of shape.
What Mobileye's map-stitching patent fixes
Cars that drive themselves lean on maps built from the recordings of other cars, and those maps are made in pieces, one region at a time. Each piece comes from different drives, so a lane line or a sign near the edge of one piece can sit a little off from the same lane line in the next piece. Lay the pieces side by side and you get a seam where the road jogs sideways.
Mobileye's filing describes a way to close that seam. It pins small control points to features such as a lane marking in each piece, links the matching pair, and then pulls the two toward each other. A built-in resistance keeps the pulling gentle, so a nudge near the seam does not drag the rest of the map out of shape.
The result is one continuous map where roads on both sides of the border line up.
… use at least one resistance value to warp a location of the first control point from an original position to a warped position based on the association between the first control point and the second control point.
Translation: It applies a resistance metric to stretch map points into alignment.
How map control points get pulled into line
The patent starts with a navigational map made of segments, each covering a different geographic region. In one version, each segment is built from drive information collected by a group of vehicles that traveled the road in that region. Control points (small anchor markers) are attached to features in each segment, and they can be laid out in a grid.
When the system decides that a feature in one segment and a feature in the next are the same real-world thing, it links their control points. It then moves a control point from its original position to a warped position, which can line it up with its partner. A resistance value (a number that sets how stiff the map is) controls how hard the map pushes back, using two simulated forces:
- Neighbor resistance: a pull between a control point and nearby points in the same segment. It grows as the points get farther apart, which keeps spacing sensible.
- Home resistance: a pull back toward the point's original position. It grows the farther the point is moved, which limits how much any one point can shift.
There is one exception. The neighbor pull can be skipped when two control points belong to different road segments, such as a road above, below, or beside another. That lets stacked roads bend independently. The updated map, with the feature at its new position, is then stored. The claims also cover a method for navigating a vehicle and the map itself.
Systems and methods for generating a navigational map are disclosed. In one implementation, a system includes a processor configured to access a navigational map including at least a first map segment representing a first geographical region and a second map segment representing a second geographical region; …
Translation: The system handles data from separate regional map segments.
Why seams in a map matter to a car
If a self-driving car checks itself against a map, a jog in the road at the edge of a map piece could throw it off. Smoothing those seams matters most on long drives that cross from one mapped region into the next. The filing says each piece is built from recordings by vehicles that drove that region, so each piece can carry its own small errors, and matching them up is a recurring chore for anyone running a crowdsourced map.
For you, the payoff would be invisible when it works: no odd lane position at a regional border. The patent covers a method for navigating a host vehicle and the map itself, so the fix lives in the data the car relies on. It is a maintenance idea, and its value depends on how well the system picks matching features.
Intel files its eighth patent we've tracked in the self-driving sensing race since September, adding to work like its autonomous pullover safety filing and Mobileye's lane paint quality map.
Stitching maps means bending them, and bending has a price. Each nudge moves a road away from where the original drives recorded it, so a seam closed in one place can introduce a small error in another. The resistance settings are the dial for that trade: stiff settings keep roads true to their recordings and leave seams, while loose settings close seams and let roads drift.
The weak spot is a wrong match. If the system decides two features are the same lane line when they are not, it will pull real roads toward the wrong target, and the claims describe no safeguard against that. The overpass exception shows the authors saw a related trap, since a road should not be dragged along with the one above it.
On balance the price looks worth paying. A small, limited distortion beats a visible jog in the road, and this is routine map housekeeping with a sensible design.
Get our take in your Top Stories
Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Intel patent news breaks.
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
55 drawing sheets from US 2026/0318692 A1 · click any drawing to enlarge
Show the other 47 sheets
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →
Be the first to weigh in