End-to-End AV Model Patents: Who Leads, Where the Gaps Are 2026
- Still accelerating. Filings ran from 85 in 2017 to a peak of 795 in 2025, with 2022's 219 marking the midpoint of that climb — the field has not plateaued.
- No single gatekeeper. The leader holds 48 records against a 2,680-record field, and the top 5 combined account for just 6.7% of all records — this is a fragmented landscape, not a walled garden.
- Interpretability sits inside a bigger AI stack. G06N (AI models) touches 42.5% of records and G06F (digital data processing) 29.0%, but A61B (diagnosis & surgery) reaching 17.2% shows how much of this corpus is shared with adjacent medical-sensing filings.
What this landscape covers
This landscape tracks 2,680 published patent records filed against a search built around end-to-end model behaviour: interpretability, long-tail scenario handling, training data mining and compute budget, paired with sensor and end-to-end terminology in the title and claims. The scope spans 2015 through the 2026-07-31 cut-off, capturing both the early exploratory filings and the recent surge as end-to-end perception-to-control architectures moved from research demonstrations toward deployable stacks.
Because publication lags filing by roughly 18 months, the most recent year in the trend chart understates real activity — 2026's count will keep rising as later filings publish. Family-level counting is used throughout so that aggressive continuation practice or multi-jurisdiction refiling in a single research programme does not inflate any one assignee's apparent share.
Filing trends and technology composition
Two views of the same 2,680-record field: how filing volume has moved year over year, and which IPC subclasses the claims actually sit in.
A decade of accelerating filings
Annual filings rose from 85 in 2017 to a peak of 795 in 2025, passing through 219 at the 2022 midpoint. The climb from midpoint to peak is steeper than from start to midpoint, which is the signature of a field still accelerating rather than levelling off; treat the 2026 figure of 370 as a partial-year floor, not a slowdown.
Where the claims concentrate
G06N (AI model computing) and G06F (digital data processing) between them touch the majority of records, but A61B (diagnosis & surgery) at 17.2% and G16H (healthcare informatics) at 12.2% both rank ahead of core vision classes like G06V and G06T. That ordering signals a search scope that pulls in sensor-and-interpretability filings from medical devices as well as road vehicles — a useful reminder to check assignee names before assuming every record is automotive.
Shares are the percentage of the 2,680 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on End-to-End Autonomous Driving Models with Eureka
This page is one run against one query. Ask Eureka your own question about end-to-end autonomous driving models and every answer comes back with the patent numbers behind it.
Try EurekaRepresentative and most-cited filings
Model interpretability information generation (US20250322276A1)
The application determines parameters for local features of a multimedia resource against a target category predicted by a first model, then builds a path for each local feature running from the feature itself to the predicted category — effectively tracing the internal decision route a model took to reach its output.Filed by Tencent Technology (Shenzhen), published 2025-10-16 — illustrates how interpretability claims are being drafted around path-tracing through model internals rather than post-hoc output explanation alone.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20210133670A1 | Control tower and enterprise management platform with a machine learning/artificial intelligence managing sen… | 337 |
| 2 | US20120321759A1 | Characterization of food materials by optomagnetic fingerprinting | 254 |
| 3 | US20220172050A1 | Method for an explainable autoencoder and an explainable generative adversarial network | 180 |
| 4 | US20230176550A1 | Quantum, biological, computer vision, and neural network systems for industrial internet of things | 176 |
| 5 | US20200159225A1 | End-To-End Interpretable Motion Planner for Autonomous Vehicles | 146 |
| 6 | US20230222454A1 | Artificial-Intelligence-Based Preventative Maintenance for Robotic Fleet | 135 |
| 7 | US20210110484A1 | Navigation Based on Liability Constraints | 133 |
| 8 | US20180114177A1 | Project entity extraction with efficient search and processing of projects | 123 |
| 9 | WO2022236064A2 | Quantum, biological, computer vision, and neural network systems for industrial internet of things | 121 |
| 10 | US20190025858A1 | Flight control using computer vision | 121 |
Citation counts are measured inside this searched corpus and favour older filings that have simply had more time to accumulate citations — read them as a signal of influence on the field's vocabulary, not of present-day importance.
Each row carries its publication number; clicking a row searches Eureka by that number.
Put your own technology through the same analysis
Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →MCP server & REST API
When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →What the data means for a filing decision
Three read-throughs from the concentration, trend and citation figures above.
Fragmentation, not lock-in
With the leader at 48 records and the top 10 combined reaching only 11.2% of the 2,680-record field, no assignee has cornered end-to-end interpretability or long-tail-scenario claims. That leaves room for a well-drafted application to stake genuinely novel ground rather than fighting through a thicket held by two or three incumbents.
The curve is still climbing
Filings moved from 85 in 2017 through 219 at the 2022 midpoint to a 2025 peak of 795 — each interval roughly tripling or more. Given the 18-month publication lag, 2026's partial count of 370 is not evidence of a slowdown; it is simply the portion that has published so far.
Medical sensing shares this claim space
A61B (diagnosis & surgery) and G16H (healthcare informatics) each cover over 12% of records in a search built around driving-relevant terms like sensor, interpretability and long-tail scenario. That overlap means some of the most-cited records — including one on optomagnetic food-material fingerprinting — sit in this corpus because the underlying sensor-interpretability language is shared across domains, not because they are automotive filings.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to end-to-end autonomous driving models, with the prior art for and against each one.
Who is filing, and who has gone quiet
Recent-year momentum diverges sharply even among named assignees in the ranking: some are accelerating fast off a small base, others have stopped filing entirely.
Academic filers moving fast off a small base
SR University's jump to 22 filings in the latest year against a small prior base shows an academic programme scaling output quickly, a pattern worth tracking for licensing or collaboration rather than blocking risk.
A pullback after earlier activity
Vellore Institute of Technology's filings dropped by roughly a quarter year over year even while remaining one of the more active academic filers, suggesting a research programme past its initial filing burst rather than one winding down entirely.
Several named filers have stopped entirely
More than one company in the ranking, including one that fell -100% year over year, recorded zero filings in the most recent year. That does not necessarily mean the programme ended — it may reflect the publication lag — but it is worth checking assignment and continuation activity before assuming those parties are still active in this space.
| Assignee | Recent year | YoY |
|---|---|---|
| SR UNIVERSITY | 22 | +175% |
| VELLORE INSITUTE OF TECH | 13 | -24% |
| Magic Leap, Inc. | 0 | — |
| Siemens AG | 0 | — |
| LUCOMM TECHNOLOGIES INC | 0 | -100% |
| Lucomm Technologies, Inc. | 0 | — |
| Mobileye Vision Technologies Ltd. | 0 | -100% |
| The Board of Trustees of the Leland Stanford Junior University | 0 | -100% |
Where to take this analysis
The figures above describe the field as a whole. Two follow-on questions determine what to do with them.
Map a specific claim against the most-cited prior art
The most-cited records in this corpus span autonomous vehicle motion planning, generative interpretability methods and even unrelated sensor domains — knowing which of those actually reads on a candidate claim takes more than a title match.
Explore prior art in EurekaTrack the assignees still accelerating
Momentum figures shift quickly in a field growing this fast; a snapshot from this page will be months old by the time a filing decision is made.
Set up assignee monitoring in EurekaCommon questions about this landscape
This landscape tracks 2,680 published patent records matching a search built around end-to-end model behaviour terms — interpretability, long-tail scenario handling, training data mining and compute budget — combined with sensor and end-to-end language, covering filings from 2015 through the 2026-07-31 data cut-off. That figure is a defined search scope, not a count of every patent touching autonomous driving broadly; a differently worded query would return a different total. Because publication lags filing by around 18 months, the true number of filed-but-not-yet-published applications from 2025 and 2026 is higher than what currently shows.
The ranking covers 100 companies, with the leader holding 48 of the 2,680 records in scope and the fifth-ranked assignee at 30. Importantly, the top 5 combined account for only 6.7% of all records, and the top 10 combined reach just 11.2% — this is a fragmented field rather than one dominated by a handful of incumbents. That fragmentation means a competitive-intelligence review needs to look well beyond the first five names to get an accurate picture of who is actually active.
Filings grew from 85 in 2017 to 219 at the 2022 midpoint and on to a peak of 795 in 2025, an accelerating rather than levelling trajectory. The 2026 figure of 370 looks lower only because it is a partial year — publication lag of roughly 18 months means recent filings are still working their way into the public record. Judged against the shape of the 2017-2025 curve, there is no visible sign of the field slowing.
The largest IPC subclass is G06N, covering AI-model computing, touching 42.5% of the 2,680 records, followed by G06F (general digital data processing) at 29.0%. Image and video-specific classes — G06V, G06T and G06K — each cover roughly 6-12% of records, smaller than the healthcare-adjacent A61B (17.2%) and G16H (12.2%) classes, which reflects how broadly the underlying sensor-and-interpretability search terms are used outside pure driving contexts. Because a single record can carry multiple IPC codes, these shares add up to well over 100% and should not be summed or treated as mutually exclusive categories.
The gap sits between the dominant AI-model classes (G06N, G06F) and the thinner-populated specific technique areas — compute-budget-aware scheduling, rare-event training data mining, and explainable path-tracing for control decisions all show comparatively low density relative to the core classes. Given that no single assignee holds more than a small fraction of the 2,680-record field, there is room to stake genuinely novel claims in these narrower technique areas rather than competing head-on in the crowded general AI-model space. A freedom-to-operate search focused specifically on the most-cited records in the target sub-area is the practical next step before drafting.
Research End-to-End Autonomous Driving Models in depth with Eureka
Go past this page: query the whole end-to-end autonomous driving models corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company’s registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.