Nvidia · Filed Jan 21, 2025 · Published Jul 23, 2026 · verified — real USPTO data

Nvidia Patents a Fix for Terrain Clipping Around High-Detail 3D Objects

When a simulator renders a detailed building on top of a generic terrain grid, the two can visually clash or overlap in ugly ways. Nvidia's new patent describes a method to automatically hide the terrain patches that would otherwise poke through high-detail objects.

Nvidia Patent: Hiding Terrain Tiles Near High-Detail 3D Objects — figure from US 2026/0212603 A1
Figure from the official USPTO publication.
Publication number US 2026/0212603 A1
Applicant NVIDIA Corporation
Filing date Jan 21, 2025
Publication date Jul 23, 2026
Inventors Kim Goossens
CPC classification 345/419
Grant likelihood Medium
Examiner WINTERS, IVAN YU (Art Unit 2611)
Status Docketed New Case - Ready for Examination (Feb 25, 2025)
Document 20 claims

How Nvidia's terrain tile system handles detail mismatches

Imagine a 3D simulation of a city where the ground is made of thousands of identical repeating tiles, like bathroom floor tiles but for virtual earth. That works fine until you drop a highly detailed, custom-built object, like a precise scan of a building or a road intersection, right on top. Suddenly the generic ground tiles and the detailed object fight for the same space, causing visual glitches.

Nvidia's patent describes a system that automatically detects which of those repeating ground tiles are sitting too close to a high-detail feature. It then hides those conflicting tiles only during the rendering step, so the detailed object can display cleanly without anything poking through it.

The system also adjusts how far away it looks for conflicts depending on how detailed the tiles are in a given area. Coarser tiles get a wider buffer zone; finer tiles get a tighter one. The result is a cleaner, more believable simulated environment without artists having to manually fix every overlap.

How the system identifies and hides conflicting mesh polygons

The patent describes a rendering pipeline step for simulation environments (think autonomous vehicle test worlds or robotics training grounds) that uses a data structure called an instanced mesh tile. These are repeating 3D patches that collectively represent a large terrain surface. They are efficient because the same mesh geometry is reused across thousands of positions rather than authored individually.

The problem arises when high-resolution features, such as photogrammetry-captured buildings, precise road markings, or custom-authored landmarks, are placed on top of that tiled terrain. The tiles underneath can visually intersect or "z-fight" with the detailed geometry.

The system works in three steps:

  • It checks which tile polygons fall within a threshold distance of any high-detail feature.
  • It adjusts that threshold based on the tile's own level of detail: low-detail tiles get a larger buffer zone, high-detail tiles get a smaller one.
  • It updates rendering parameters for the affected polygons so they are omitted from the render pass, hiding them without permanently deleting the underlying data.

Critically, the tile data itself is not destroyed. The system only toggles a parameter that tells the renderer to skip those polygons, which makes the operation fast and reversible.

What this means for simulation and autonomous vehicle testing

Nvidia's simulation platforms, particularly Omniverse and its autonomous vehicle training suite, depend on generating large, visually consistent virtual worlds at scale. When terrain tiles and detailed assets collide, engineers historically have had to resolve conflicts by hand or accept visual artifacts. This patent describes an automatic, parameter-based solution that scales with the size of the scene and the variety of detail levels in it.

For the people actually using these simulation systems, whether they are training self-driving car models or building digital twins of real-world locations, this kind of fix means cleaner training data. A simulator that renders terrain incorrectly can introduce errors into the perception models trained on that data. Getting the ground to look right is less glamorous than headline AI features, but it is the kind of foundational work that determines whether simulation results transfer to the real world.

Editorial take

This is unglamorous but genuinely useful infrastructure work. Terrain-tile popping and z-fighting are real, well-known problems in large-scale 3D simulation, and an automatic, level-of-detail-aware solution is meaningfully better than the status quo of manual cleanup. It won't headline a product launch, but it's exactly the kind of patent that improves simulation fidelity across Nvidia's entire platform.

Which company should we read for you?

We track 17 companies here. Pro is the same weekly breakdown for any company you choose, delivered privately. Type a name and we'll scope it and send you a quote.

Get one Big Tech patent every Sunday

Plain English, intelligent commentary, no hype. Free.

Source. Full patent text and figures from the official USPTO publication PDF.

Editorial commentary on a publicly published patent application. Not legal advice.