US 10816354

I'll research US patent 10816354 across patent databases and litigation dockets.

Let me search for litigation and docket information specifically.

Let me check whether this patent is specifically asserted in litigation and whether there's a CAFC appeal.

US Patent 10,816,354 — Research Summary

Important scope note: I searched patent databases (Google Patents, Unified Patents) and litigation dockets (Justia, Texas courts, VitalLaw). I did not find any Court of Appeals for the Federal Circuit docket involving US 10,816,354. The only appellate activity I found near this technology is a Texas state-court trade-secret appeal, not a CAFC patent appeal. I flag this explicitly rather than inferring CAFC involvement. Also note: results returned below include material dated after the "April 26, 2026" date in your prompt but before the system date (2026-09-29); I report what the sources show.

Bibliographic data (per Google Patents, fetched 2026-09-24)

Continuation/child applications in the same family: US 17/074,468 (→ US 11,573,095 B2), US 18/158,974 (→ US 11,874,130 B2), US 18/449,340 (→ US 12,276,516 B2).

Abstract (verbatim substance)

A method of lane detection for a non-transitory computer readable storage medium storing one or more programs. The programs cause a computing device to: generate a ground truth associated with lane markings expressed in god's view; receive features from at least one of a hit-map image and a fitted lane marking (the hit-map image including a classification of pixels that hit a lane marking, and the fitted lane marking including pixels optimized based on the hit-map image); and train a confidence module based on the features and the ground truth, the confidence module configured to determine on-line whether a fitted lane marking is reasonable, using parameters that express a lane marking in an arc.

Plain-language overview of the independent claims

Claim 1 (method):

1. Receive road images containing lane markings.
2. Build a "ground truth" by annotating those images to identify the lane markings; the ground truth is used to train a lane-detection algorithm.
3. Receive a hit-map image for the current view that identifies which pixels hit a lane marking.
4. Receive a first lane template from the previous view of the road.
5. Generate a fitted lane marking using both the previous-view template and the hit-map.
6. Train a confidence module using the ground truth plus either the hit-map or the fitted lane marking, where the confidence module judges the fitted lane marking's confidence level using parameters of a circle arc fitted into that lane marking.

Claim 10 (system): Mirrors claim 1 as an apparatus claim — a memory, one or more processing units, and stored programs that perform the same receiving-annotating-hit-map-template-fitting-confidence-training sequence.

Representative dependent claims: claim 2/12 (annotation uses arcs of circles); 3/13 (train the lane-detection algorithm on the ground truth and generate another hit-map); 4/14 (adjust the prior template per rules; claim 14 recites explicit constraints — parallel lane markings, circular curvature, curvature < ~300 m, lane spacing ~3–4 m, edge color differing from non-marking road); 5/15 (confidence ≥ threshold → output as predicted lane marking); 6/16 (compare the accepted marking to ground truth → identify a failure pattern); 7/17 (confidence below threshold → reject); 8/18 (compare the rejected marking to ground truth → failure pattern); 9/19 (refine confidence module with added labeled data); 11 (build a second lane template for the next view); 20 (defines "god's view": vehicle at origin, y-axis fore–aft, x-axis perpendicular, i.e., a top-down bird's-eye coordinate frame).

Litigation findings (as of the sources retrieved)

• CreateAI, Inc. v. Bot Auto, Inc. et al., No. 4:25-cv-03371 (S.D. Tex.) — patent infringement suit filed 2025-07-21, Judge Charles Eskridge. Docket shows Bot Auto's motion to dismiss (filed 2025-09-22) and, per a CreateAI/CreateAI financial disclosure, that motion was denied on 2026-02-18; case is in discovery, trial date TBD. Note that an AI-generated complaint analysis (Ex Parte) identifies US 11,577,783 and US 12,190,609 as the asserted patents — I did not find confirmation that US 10,816,354 is asserted in this case. (Also, the Justia docket lists the cause of action literally as "35 U.S.C. § 145," which is a docketing characterization I am reporting as-is; § 145 is not the usual infringement provision.)
• CreateAI Holdings, Inc. v. Bot Auto TX, Inc., No. 24-BC11A-00007 (Tex. Bus. Ct., Harris Cty.) — trade-secret case (TUTSA), unrelated to this patent's claims. Interlocutory appeal to the Fifteenth Court of Appeals of Texas (No. 15-25-00001-CV); the appeals court affirmed denial of a temporary injunction (VitalLaw report dated 2026-09-17). Note this is a state appellate court, not the CAFC.
• Context: inventor Xiaodi Hou is TuSimple's former CEO/CTO and founder of Bot Auto, which is why these disputes overlap.

Uncertainty / caveats

• I could not retrieve anything from the USPTO PatentCenter or PACER directly (paywalled); bibliographic data comes from Google Patents, and litigation data from Justia/Ex Parte/Texas courts/VitalLaw.
• I found no CAFC 2026 (or other) docket for US 10,816,354. If a Federal Circuit appeal exists, it was not surfaced by these searches and I cannot confirm it.
• Whether US 10,816,354 is actually asserted in 4:25-cv-03371 is unconfirmed.
• Assignee records have changed over time (TuSimple → TuSimple, Inc. → CreateAI, Inc.); the Google Patents "current assignee" line should be treated as a database entry, not a legal conclusion.

I'll search for litigation involving US patent 10816354.

Let me search more specifically for TuSimple litigation and the Unified Patents case list.

Let me dig into the Bot Auto patent case and search for the specific patent in litigation databases.

Summary finding

Based on my searches, I could not confirm any patent litigation in which U.S. Patent No. 10,816,354 ("the '354 patent") has been expressly asserted or otherwise at issue. I also could not confirm any PTAB proceeding (IPR/PGR), ITC investigation, or CAFC appeal naming the '354 patent.

I want to be explicit about two caveats: (1) I hit tool/search limits before I could exhaustively verify the patent-by-patent listings in the most relevant case, and (2) litigation databases (Unified Patents Portal, Docket Navigator, PACER, CourtListener) were not directly interrogable through my searches. So this should be treated as "no confirmed litigation found," not as an affirmative certification that none exists.

The closest known litigation (patent and otherwise), with the caveat noted

The '354 patent is owned by TuSimple, Inc. (originally assigned to TuSimple Inc.; current assignee listed as CreateAI Inc. per Google Patents). The following TuSimple/CreateAI cases exist, but I could not verify that the '354 patent is among the asserted patents in the one patent case.

1. TuSimple, Inc. v. Bot Auto, Inc. et al. — patent case (patent asserted: unverified)

• Plaintiff: TuSimple, Inc. (per docket; CreateAI Holdings, Inc. also appears as plaintiff in later filings)
• Defendants: Bot Auto, Inc.; Bot Auto TX, Inc.; Bot Auto CA, Inc.
• Jurisdiction/Court: U.S. District Court for the Southern District of Texas (Judge Charles Eskridge)
• Case No.: 4:25-cv-03371
• Filing date: July 21, 2025
• Nature of suit: Patent — 35 U.S.C. §145 (as listed on docket)
• Status (as of the last docket retrieval, Jan. 16, 2026): Motion to dismiss (Dkt. 11) filed Sept. 22, 2025; opposition Oct. 6, 2025; reply Oct. 16, 2025; motion to stay pending ruling on the motion to dismiss filed Jan. 16, 2026. Initial scheduling conference set for Feb. 18, 2026.
• Reported asserted patents: A third-party complaint analysis identifies U.S. Patent No. 11,577,776 ("redundant steering"), U.S. Patent No. 12,190,609 ("map construction"), and a '783 patent. I did not find U.S. 10,816,354 in the asserted set. Given the '354 patent is a TuSimple lane-detection patent and the accused products are autonomous trucks, it is plausible but unconfirmed that it may also be asserted; this must be verified against the complaint (Dkt. 1) and its exhibits.
• Sources: https://dockets.justia.com/docket/texas/txsdce/4:2025cv03371/[2019667](/patent/2019667) ; https://ai-lab.exparte.com/case/dct/txsd/4:25-cv-03371/doc/analysis/1

2. CreateAI Holdings, Inc. (f/k/a TuSimple Holdings, Inc.) v. Bot Auto TX Inc. — trade-secret case (not a patent case)

• Court: Texas Business Court, Harris County, 11th Division; Trial Court Cause No. 24-BC11A-0007
• Appeal: No. 15-25-00001-CV, Fifteenth Court of Appeals
• Claim: Texas Uniform Trade Secrets Act (TUTSA) — misappropriation of AV technology
• Status: Temporary injunction denied and TRO dissolved; affirmed by the 15th Court of Appeals (Sept. 15, 2026).
• Sources: https://tcjl.com/15th-court-of-appeals-affirms-business-court-denial-of-ti-in-misappropriation-of-trade-secrets-case/ ; Texas courts brief: https://search.txcourts.gov/SearchMedia.aspx?MediaVersionID=33f0d9a5-effe-494c-8310-532c7fdf957b&coa=coa15&DT=Brief&MediaID=0aa5cc73-1b18-4adc-8c53-11ff0973bca9

3. Wilhoite, et al. v. Hou, et al. — shareholder derivative / trade-secret case (not a patent case)

• Court: U.S. District Court for the Southern District of California, Case No. 3:23-cv-02333-BEN-MSB
• Filed Dec. 22, 2023; settled for ~$42.5M; final approval July 23, 2025; settlement appealed to the Ninth Circuit (Aug. 19, 2025, by Xiaodi Hou), appeal pending.
• Source: https://www.blbglaw.com/cases-investigations/tusimple

4. Delaware Court of Chancery shareholder derivative actions (Nusbaum v. Hou, 2022-1095; Young v. Hou, 2022-1157; Wolfson v. Hou, 2023-0279; later Camac Fund, L.P. v. Chen et al., 2025-0181-PAF) — fiduciary-duty and related-party claims, not patent claims.

• Source: CreateAI 3Q2025 financials, https://s202.q4cdn.com/364265561/files/doc_financials/2025/q3/CreateAI-3Q2025-Financials-Print-Ready-Version.pdf

Important technical/legal context about the '354 patent

• Patent family: The '354 patent is one of four U.S. family members (Family ID 65435004): US10816354B2 (this patent), US11573095B2 (continuation, filed 10/19/2020), US11874130B2 (filed 1/24/2023), and US12276516B2 (filed 8/14/2023). Foreign family members include EP3673340A4, EP4553599A1, WO2019040670A1, and CN111052132B.
• Prosecution/ownership history: Priority and filing date Aug. 22, 2017; granted Oct. 27, 2020; adjusted expiration listed as Feb. 14, 2038; status "Active." Originally assigned to TuSimple; later changed to CreateAI, Inc.
• No adversarial record visible on the face of the patent: The Google Patents record for US10816354B2 shows "Cited By" and "Families Citing this family" entries (e.g., UATC L.L.C. lane-boundary patents, various CN/EP/DE family citations) but no litigation, PTAB, or ITC entries, which is consistent with (though not proof of) the absence of an asserted patent dispute involving this specific patent.

What I could not do within my search budget (recommended next steps)

To definitively confirm or rule out litigation on the '354 patent, these checks should be run directly:

1. Unified Patents Portal case list (https://portal.unifiedpatents.com/litigation/caselist) — search by patent number 10,816,354.
2. PACER / Docket Navigator / CourtListener RECAP — full-text search for "10,816,354" (and "10816354") across district courts, and confirm the asserted patent list in N.D./S.D. Tex. and other venues.
3. USPTO Patent Center / PTAB — confirm no IPR/PGR naming the '354 patent.
4. The Bot Auto complaint (Dkt. 1, S.D. Tex. 4:25-cv-03371) — verify whether the '354 patent appears among the patents-in-suit; this is the single most likely place it would be asserted given the parties.
5. CAFC docket — check for any appeal referencing the '354 patent.

Bottom line: On the available record, no litigation specifically involving U.S. Patent No. 10,816,354 was confirmed. The active TuSimple/CreateAI litigation against Bot Auto is a genuine patent case, but I could not verify that the '354 patent is one of the asserted patents, and the '354 patent does not appear to have any litigation, PTAB, or ITC entries associated with it.

PTAB Proceedings — US 10,816,354 B2

"Verification Module System and Method for Motion-Based Lane Detection with Multiple Sensors"
Patent Owner: TuSimple, Inc. (now CreateAI, Inc. per assignment records) · App. No. 15/683,441 · Filed 2017-08-22 · Granted 2020-10-27 · Adjusted expiration 2038-02-14 · Inventors: Siyuan Liu, Mingdong Wang, Xiaodi Hou

Proceedings overview

Total AIA trial proceedings on file: 0. The structured USPTO Open Data Portal (ODP) block supplied for this analysis returns no AIA trial proceedings (no IPR, PGR, or CBM) for US 10,816,354 as of the most recent ODP ingest — breakdown: 0 active / 0 claims invalidated / 0 claims sustained / 0 settled / 0 institution denials. All 20 claims stand untested at the Board. The bottom-line defensive posture is the opposite of "hardened": there is no petitioner-side win to point to, but also no PTAB estoppel and no adverse FWD constraining a defendant — you would be the first mover, and the entire § 102/§ 103 record against these claims is unbuilt.

⚠️ Caveat on the search: My independent web checks (Google Patents family/PTAB data, PTAB E2E document hits via ptacts.uspto.gov, and targeted queries for TuSimple IPR/PGR activity and for "10,816,354" as a challenged patent) surfaced no petition, institution decision, FWD, or appeal referencing this patent. The only ptacts.uspto.gov hits returned were unrelated proceedings on other patents (e.g., IPR2022-01486 on Varta's 9,496,581; IPR2022-00124 on Lennon Image's 6,624,843), which I am not attributing to this patent. I found no evidence of any AIA trial — but note that ODP/PTAB E2E indexing lags for recently filed petitions, so treat "zero" as "zero on record," not as a metaphysical certainty. Verify directly in PTAB E2E before relying on this in a litigation posture: https://ptacts.uspto.gov/ptacts/public-informations/ (search by patent number) and confirm against the patent's Google Patents page: https://patents.google.com/patent/[US10816354B2](/patent/US10816354B2)/en

Per-proceeding detail

No proceedings to detail. There are no IPR/PGR/CBM dockets, no petitioner names, no institution decisions, no Final Written Decisions, no terminations, and no Federal Circuit appeals to report for US 10,816,354. I will not generate placeholder proceeding numbers or hypothetical panels — per your instruction, I do not invent proceeding identifiers, and I will not fabricate claim-level dispositions for proceedings that do not exist.

Strategic summary

Claim status — the whole patent is UNTESTED. Nothing is canceled, nothing is confirmed by the Board, nothing is statutorily disclaimed on the public record by any AIA trial outcome. Claim 1 (method) and claim 10 (system) are the independent claims; the balance (claims 2–9, 11–20) are dependents, with claim 11 oddly drafted as a method claim depending from claim 1, claim 14 incorporating an explicit five-element constraint list (parallel lane markings; circular curvature shape; curvature < ~300 m; lane spacing ~3–4 m; edge color differing from non-marking road), and claim 20 reciting the "god's view" coordinate convention (vehicle origin, y-axis front-to-back, x-axis side-to-side, perpendicular axes). Because there is no FWD, a defendant cannot say "claims 1–5 are dead" — the opposite is true: the patent owner has a clean, fully intact claim set, and the burden and cost of building invalidity (including any IPR) sits entirely on the defendant.

Estoppel landscape — a blank slate. With no IPR/PGR, there is no § 315(e)(2) estoppel binding anyone: no petitioner is barred, no privity chain exists, and no ground was "raised or reasonably could have been raised." Practically, this means a defendant has maximum freedom of action — you may petition on any § 102/§ 103 ground with prior art of your choosing, and, importantly, there is no prior PTAB record for the patent owner to mine for prosecution-style admissions or claim-construction positions to reuse against you. The flip side: you also get no benefit from someone else's work product, no institution decision to copy, and no expert report already on file.

Pattern signals. (a) Same-petitioner serial filings: none — no petitioner has ever filed. (b) Patent-owner appeal aggressiveness: no PTAB appeals on this patent; there is nothing to appeal. (c) Defensive aggregator: I found no Unified Patents (or similar aggregator) activity against this patent — a Unified Patents portal page surfaced only for an unrelated TuSimple lane-marking-localization publication, not for 10,816,354. Combined with the ODP "no proceedings" flag, the absence of aggregator activity is a mild signal that this patent has not been the target of a broad NPE-sweep campaign; well-asserted, widely-licensed patents typically attract at least one aggregator or competitor IPR within a few years of grant, and this one has been granted since 2020-10-27.

Family context that a defendant must not miss. This patent is the earliest US filing in a four-member US family sharing the 2017-08-22 priority date: US 10,816,354 B2 (this patent), plus continuations US 11,573,095 B2 (filed 2020-10-19), US 11,874,130 B2 (filed 2023-01-24), and US 12,276,516 B2 (filed 2023-08-14), with EP (EP3673340, EP4553599), CN 111052132 B, and WO2019040670A1 counterparts. This matters for two reasons: (1) a validity attack aimed only at 10,816,354 leaves three continuations live on substantially the same disclosure, so freedom-to-operate analysis must cover the whole family; and (2) because each continuation is separately challengeable and the family is still being prosecuted/extended, the patent owner retains the ability to draft new claims targeted at an accused product's architecture — a structural risk that no single IPR on the '354 patent can neutralize. Note also that the patent's own specification and the related filings frame the invention around "god's view" lane detection with IMU/GPS/MAP priors; the '354 claims themselves (per the claim set in the record) are notably narrower than the specification's broad disclosure, which is where invalidity leverage typically lives.

Why the zero might be real (and what it says). TuSimple's public portfolio materials claim 500+ global patents; the company reorganized and its patent assets were reassigned to CreateAI, Inc. in 2025. A patent that has never been asserted in a campaign, held by an assignee in transition, and granted only in late 2020 is unremarkable as an IPR target. So the zero is plausible rather than an indexing artifact — but "not yet asserted" is the operative phrase, and that can change with a single demand letter.

Recommended next steps

If you are a defendant facing a demand letter citing US 10,816,354:

1. Do not assume the record is empty in the courthouse sense — absence of PTAB proceedings is not absence of litigation. Run a district-court docket check (Docket Alarm / CourtListener: https://www.courtlistener.com) on 15/683,441 and the family members, and confirm PTAB E2E directly (https://ptacts.uspto.gov/ptacts/public-informations/). Nothing I found establishes that this patent has been asserted — I found no assertion either.
2. You are the first mover on this patent. With no FWD and no estoppel, an IPR/PGR is procedurally clean: file within the § 315(b) one-year window from service of any complaint, and you may raise any § 102/§ 103 ground (and, for a PGR-eligible claim, § 112 — though note PGR is time-barred here: PGR must be filed within 9 months of grant, and this patent granted 2020-10-27, so only IPR is available). Statutory deadline to remember: PTAB must issue a Final Written Decision within 1 year of institution (§ 316(a)(11)), extendable up to 6 months for good cause.
3. Because there is no FWD, there is no claim-level disposition to quote. If you are evaluating the patent, build your own claim chart against claim 1 and claim 10 — claim 14's five enumerated constraints and claim 20's coordinate-system recitation are the most concrete structures to attack or design around, and the "confidence module … using parameters of an arc of a circle fitted into the fitted lane marking" limitation (present in both independent claims) is the natural § 103 hook against arc-fitting/curve-fitting lane-detection art.
4. Treat the continuations as co-extensive risk. Budget for US 11,573,095 / US 11,874,130 / US 12,276,516 and EP 3 673 340 / EP 4 553 599 — and, if the FTO is for the European market, note that the EPO family members are separate validity battlegrounds entirely outside the PTAB's reach.
5. Monitor. No proceeding is active, so there are no institution-decision deadlines, oral hearing dates, or FWD due dates to track. The trigger events to watch are: (a) a first IPR/PGR petition on this patent, (b) any assignment-driven assertion by CreateAI, Inc., and (c) new continuation claims emerging from US 12,276,516 or later filings, which would give the owner fresh, unchallenged claim scope.

I'll research the assignment chain for US 10,816,354. Let me start with several searches in parallel.

Let me dig for specific reel/frame records and the assignee-event history.

Assignment & Ownership Reconstruction — US 10,816,354 B2

Source limitation (read first). I could not load the USPTO Assignment Center (assignmentcenter.uspto.gov) or the legacy assignment search (assignment.uspto.gov/patent/index.html) directly in this session — both are JavaScript-gated search interfaces that did not return records to my tools. No reel/frame number and no correspondent-of-record entry surfaced in any indexed source I could reach (including third-party mirrors such as uspto.report and Unified Patents' portal). The timeline below is therefore reconstructed from the Google Patents legal-events ledger, which is generated from the same underlying USPTO recordation events, plus primary corporate documents. I flag every field I could not verify rather than inventing it. This is a material gap: the correspondent field is the single most diagnostic element of this task and I do not have it.

Inventors

All three named inventors share the same filing address (San Diego, CA), consistent with employment at the then-founding-stage assignee:

Unusual patterns — assessed:

• No "all inventors gone within 12 months" pattern. The application was filed 2017-08-22 and the inventors appear to have remained with the assignee through the 2020 grant; Hou's separation was ~5 years later (fired as CEO Oct 2022; resigned from the board Mar 2023). The "abandoned-portfolio / fire-sale precursor" tell is not present.
• Founder-departure-then-compete pattern (context, not an assignment finding). Hou went on to found Bot Auto, which is the defendant in the assignee's current litigation. Per TuSimple's own Dec 2024 statement, the company alleges Bot Auto was founded while Hou was still a director and misappropriated sensor/perception trade secrets. This is the reason the ownership chain and the litigation are intertwined, and it is why I did not treat the assignee's plaintiff status as a generic "NPE campaign" signal (see Signal 2).
• Note: an unrelated third-party data source (Unified Patents' page for the Chinese family member CN-111052132-A) lists the priority date as 2017-08-21; the US record is 2017-08-22. That is a time-zone artifact in the CN filing, not a second priority claim.

Original assignee

TuSimple, Inc. (San Diego, CA) — named on the face of the issued patent as "(73) Assignee: TUSIMPLE, INC., San Diego, CA (US)."

• Primary line of business: autonomous trucking (SAE L4) — perception, mapping, and vehicle control, with an Autonomous Freight Network and, from Apr 2023, a proprietary "TuSimple Domain Controller" central compute unit. The claims (motion-based lane detection fusing camera/LiDAR with IMU/GPS/HD-map priors) are directly on the critical path of that product.
• Did it ship a product embodying the claims? Substantially yes during the relevant period — the assignee publicly described operating autonomous trucks and a production-intent domain controller. I could not independently verify that a shipped unit literally practiced every limitation of claims 1/10.
• Current status: Reincorporated/rebranded, still operating, not in bankruptcy. Timeline: US operations wind-down announced Dec 2023; Nasdaq delisting Jan/Feb 2024; rebrand to CreateAI announced Dec 18–19, 2024; OTC Markets filings describe CreateAI as "a global digital interactive entertainment company focused on developing video games and animation products" that "also has autonomous driving technology and data which it will continue to monetize." An OTC Markets financial report (undated in the retrieved excerpt; consistent with 2025) still lists "2100 Eagle Parkway, Fort Worth" as the rebranded company's address.
• Chain of corporate name: TuSimple → TuSimple, Inc. → TUSIMPLE, INC. → CreateAI, Inc. The USPTO's 2025-12-03 "Change of Name" recordation lags the Dec 2024 public rebrand by roughly 12 months — a normal administrative lag, not evidence of a concealed transfer, but worth noting because it means the registered owner of record changed after the assignee filed suit in July 2025.
• Public-company identifiers for cross-reference: TuSimple Holdings, Inc., SEC CIK 1823593 (ex-Nasdaq: TSP; now OTCMKTS: TSPH). The '354 patent's record owner is the operating subsidiary, not necessarily the holdco.

Assignment timeline

There are post-filing records — but all post-2017 records are Change of Name only. Ownership never economically left the original corporate family. I could not retrieve reel/frame for any entry.

• Executed: not retrieved (presumed on/about 2017-08-22) / recorded on or about 2017-08-22

  • Conveyance: Assignment of assignors' interest ("ASSIGNMENT OF ASSIGNORS INTEREST")
  • Assignor: Siyuan Liu; Mingdong Wang; Xiaodi Hou (individuals)
  • Assignee: TuSimple, Inc. (San Diego, CA) — listed in the event ledger as "TuSimple"
  • Correspondent: not retrieved — could not be determined from accessible sources
  • Context: Original inventor-to-company employment assignment at a venture-stage company.
  • Reel/frame: not retrieved.

• Executed: not retrieved / recorded 2020-01-30

  • Conveyance: Change of Name
  • Assignor: TuSimple ("TuSimple, Inc.")
  • Assignee: TUSIMPLE, INC.
  • Correspondent: not retrieved
  • Context: Change of name only — no change in beneficial ownership; consistent with the pre-IPO corporate re-organization. (I note a possible inconsistency: the 2019 pre-grant publication and the 2020 printed patent both render the assignee as "TuSimple, Inc.," so the "change of name" here may be a formal name normalization rather than a substantive rename. Cannot resolve without the recorded document.)
  • Reel/frame: not retrieved.

• Executed: not retrieved / recorded 2025-12-03

  • Conveyance: Change of Name
  • Assignor: TUSIMPLE, INC.
  • Assignee: CreateAI, Inc.
  • Correspondent: not retrieved
  • Context: Change of name only — reflects the Dec 2024 TuSimple → CreateAI rebrand. This is the last recorded event; there is no transfer to a licensing LLC, no security agreement, and no release in the chain.

Family note (ownership-relevant but not separate records for this patent): the same corporate chain covers the three asserted continuations in the family — US 17/074,468 → US 11,573,095 B2; US 18/158,974 → US 11,874,130 B2; US 18/449,340 → US 12,276,516 B2. Each will carry its own recordation events, which I did not retrieve. Do not assume they are identical to this patent's record without checking.

Timeline diagram

timeline
    title Ownership of US 10816354
    2017 : Filed by TuSimple Inc
         : Inventors assign rights to TuSimple
    2020 : Patent issued as US 10816354
         : Change of name to TUSIMPLE INC
    2024 : TuSimple rebrands to CreateAI
    2025 : Change of name to CreateAI Inc

NPE / troll-pattern signals

1. Shell-entity transfer — not present. Both post-filing records are Change of Name (recorded 2020-01-30 and 2025-12-03). The terminal assignee, CreateAI, Inc., is an operating digital-entertainment company with employees and products (OTC Markets financial report; PRNewswire, 2024-12-18). No "IP / Licensing / Holdings / Ventures" LLC appears anywhere in the chain, and no registered-agent-service address is implicated.

2. Known asserter in the chain — not present. Neither TuSimple, Inc. nor CreateAI, Inc. matches any entity on the Acacia / Marathon / Intellectual Ventures / IPNav / Wi-LAN / Conversant / Vringo / Pendrell / Innovatio / MPHJ / Lumen View / Round Rock / Document Generation Corp / Spangenberg lists. The assignee is a plaintiff — CreateAI, Inc. v. Bot Auto, Inc., No. 4:25-cv-03371 (S.D. Tex., filed 2025-07-21) — but that action is a dispute with a competitor founded by the patent's own co-inventor, which is the classic operating-company-vs-competitor posture, not an NPE assertion campaign. Critical caveat: the asserted patents in that case are reported to be US 11,577,783 and US 12,190,609; I found no confirmation that US 10,816,354 is asserted. This signal therefore cannot be upgraded on the basis of the suit.

3. Repeat correspondent across the chain — unclear. I could not retrieve the correspondent of record for any of the three entries; the search interfaces that expose that field were not reachable. This is the single highest-value missing datum — if the 2020-01-30 and 2025-12-03 records share a correspondent with the 2017-08-22 record, that is ordinary in-house/outside-counsel continuity; if a single outside firm appears across many unrelated NPE chains, that is the tell. On the record available, no recurrence can be shown or excluded.

4. Cascading transfers — not present. Two records across ~8 years (2020-01-30 and 2025-12-03), both renames. There is no sequence of chained LLCs and no <24-month cascade.

5. Pre-litigation transfer — not present. The only recordation near the litigation is 2025-12-03, which post-dates the 2025-07-21 complaint by ~4.5 months. Nothing was transferred to the plaintiff in the six months before filing; the plaintiff (under its new name) already owned the patent via the 2020 change of name.

6. Bankruptcy fire-sale — not present. No Chapter 7/11 proceeding and no sale of the portfolio in an insolvency proceeding was found. Context worth flagging: the assignee was in genuine financial distress — US operations wind-down (Dec 2023), Nasdaq delisting (Jan/Feb 2024), shareholder litigation, and a Dec 2024 letter from co-inventor Hou seeking full liquidation of the company. That combination often precedes a portfolio sale, so the "not present" call is a current-state finding, not a guarantee; a future assignment to an unfamiliar buyer would change this signal quickly.

7. Privateering — not present. No operating-company-to-NPE transfer appears; the assignee is litigating directly, in its own name, against a competitor.

8. Defensive aggregator — not present. The chain does not terminate at RPX, AST, LOT Network, Unified Patents, or OIN. (US 10,816,354's family appears in Unified Patents' public database via CN-111052132-A, but that is database coverage, not a transfer — it is not a finding.)

Verdict

Insufficient data — and, on the evidence available, affirmatively not an NPE chain.

Justification: the entire recorded chain is the original 2017-08-22 inventor-to-TuSimple assignment plus two Change of Name recordations (recorded 2020-01-30 and 2025-12-03); ownership never economically left the original corporate family, no licensing-only entity, security interest, or cascading LLC transfer appears, and the only litigation involving the assignee is a competitor/founder dispute in which this patent's assertion is unconfirmed. I stop at "Insufficient data" rather than "Operating-company assertion" for two reasons: (i) I could not verify the reel/frame or correspondent for any record, so the diagnostic core of the task is unresolved; and (ii) CreateAI has pivoted away from autonomous trucking, so "ships products embodying the claims" is doubtful for the current owner. If US 10,816,354 is later confirmed as an asserted patent in CreateAI, Inc. v. Bot Auto, Inc., reclassify to Operating-company assertion.

Verification entry points: USPTO Assignment Center — https://assignmentcenter.uspto.gov/ ; legacy patent assignment search — https://assignment.uspto.gov/patent/index.html (search "10816354"). The three fields to pull when the interface is reachable: reel/frame for each of the three events, the correspondent on each, and the execution date on the 2025-12-03 change of name (to test whether it was executed before or after the 2025-07-21 complaint).

I'll research the USPTO records and prior-art citations for US 10,816,354. Let me run targeted searches.

I have partial front-page data. Let me dig for the complete "References Cited" list on the patent's face and the international search report for the family.

Research Note: Prior Art / Cited References for US Patent 10,816,354

Search target confirmed. The search resolves to exactly one patent: US 10,816,354 B2, title "Verification module system and method for motion-based lane detection with multiple sensors," Application No. US 15/683,441, filed 2017-08-22, granted 2020-10-27, inventors Siyuan Liu, Mingdong Wang, Xiaodi Hou, original assignee TuSimple, Inc. (current assignee listed as CreateAI Inc.). Pre-grant publication: US 2019/0063945 A1 (2019-02-28). I did not substitute any similar numbers.

⚠️ Critical source limitation (read before using the analysis)

The authoritative full text you supplied — the Google Patents capture — does not contain the "References Cited" / "Patent Citations" (backward-citation) list at all. It only reproduces the "Cited By" table (forward citations, i.e., later patents that cite '354) and the family/priority data. Therefore, the only place the examiner-cited prior art appears is the granted-patent front page (item (56)) and the PCT International Search Report for the family (PCT/US2018/047608 → WO 2019/040670 A1).

I was able to retrieve only a partial image of the '354 front page (from the patentimages.googleapis.com PDF of US010816354B2), and I could not retrieve the complete (56) list, the "Continued" continuation entries, or the ISR. I am reporting exactly what the retrieved sources show and explicitly flagging what I could not verify. I will not invent citations.

Part A — References cited on the face of US 10,816,354 (partial, verified from the front-page image)

1. U.S. Patent Documents

2. Foreign Patent Documents

3. Other Publications (NPL)

Part B — Per-reference analysis (citation · description · potential § 102 relevance)

Framing for § 102: Because '354 has a filing/priority date of 2017-08-22, only art that is (a) patented/published/on-sale before that date, or (b) an earlier-filed, later-published U.S. application (AIA § 102(a)(2)), can be § 102 prior art. All three U.S. references (1999–2003) and the Bar Hillel survey predate it by many years; the two CN publications (Jan/May 2017) predate it by months. Note: a reference appearing on the examiner's face list is not automatically § 102 art — examiners routinely cite art under § 103 or as general background.

1. US 6,084,870 A — Wooten et al., issued 2000-07-04.

• Description: I could not verify the subject matter of this reference from the retrieved sources. I will not supply a fabricated title or abstract. Based purely on its position and era, it is a pre-2000 U.S. patent in the vision/navigation space.
• Potential § 102 mapping: Cannot responsibly assign. No element-by-element comparison is possible without the reference's disclosure.

2. US 6,263,088 B1 — Crabtree et al., issued 2001-07-17.

• Description: My training recollection associates this family with redundant multi-camera lane tracking for vehicles, but I could not verify that from the retrieved sources, so treat it as unconfirmed.
• Potential § 102 mapping (tentative): If it is a multi-sensor lane-tracking disclosure, it is most plausibly relevant to the multi-sensor aspects of claim 1/10 (receiving images of a road with lane markings) and to claim 14 constraints (lane spacing, road geometry). It is very unlikely to disclose the hit-map image, the previous-view lane template, or the "confidence module trained … using parameters of an arc of a circle fitted into the fitted lane marking" — the core of the independent claims. So: possible § 102/§ 103 pertinence to sub-claims, not anticipation of claim 1 or 10.

3. US 6,594,821 B1 — Banning et al., issued 2003-07-15.

• Description: Same caveat — subject matter not verified from the retrieved sources.
• Potential § 102 mapping (tentative): Older image-based highway/lane perception patents of this era commonly disclose fitting lane boundaries to curves/arcs from camera images, which could bear on claim 2/12 ("annotating the one or more lane markings using arcs of circles") and the arc-fitting limitation of claim 1/10. Again, unlikely to anticipate the full independent claims; more likely § 103 fodder for the arc-parameter element. Unverified.

4. CN 106340197 A — published 2017-01.

• Description: Not verified from retrieved sources. A January-2017 Chinese published application — contemporaneous with the 2016–2017 CNN-lane-detection wave.
• Potential § 102 mapping: As a published foreign application before 2017-08-22, it qualifies as prior art under § 102(a)(1) (or § 102(a)(2) only if there is a U.S. counterpart). Recent CNN-based lane-detection applications of this vintage are the most likely of the cited references to disclose a "hit-map"/pixel-classification lane output, making it the strongest candidate to bear on claims 1/10 and 3/13 (training a lane-detection algorithm and generating a hit-map). But I could not verify its disclosure; do not rely on this until the document is pulled.

5. CN 106781591 A — published 2017-05.

• Description: Not verified from retrieved sources. May-2017 Chinese published application.
• Potential § 102 mapping: Same framework as the above — § 102(a)(1) art as of 2017-08-22. Potentially relevant to the lane-detection/hit-map and curve-fitting elements of claims 1/10, 2/12, 3/13. Unverified.

6. Bar Hillel et al., "Recent Progress in Road and Lane Detection — A survey" (NPL).

• Description: A published survey of road/lane detection. Surveys are printed publications qualifying under § 102(a)(1), but a survey is generally cited to show the state of the art / general knowledge, not to anticipate.
• Potential § 102 mapping: Cannot anticipate any claim standing alone; at most § 103 background.

7. Possible additional reference (flagged, not confirmed): US 9,672,446 (Vallespi-Gonzalez), issued 2017-06-06.

• A Justia citation table surfaced adjacent to the '354 family appears to list US 9,672,446 | June 6, 2017 | Vallespi-Gonzalez alongside US 10,816,354 | Oct 27, 2020 | Liu. I could not determine whether US 9,672,446 is a formal (56) reference cited by '354 or is merely art cited in a downstream continuation (e.g., US 12,276,516). Treat as unconfirmed. If it is a '354 citation, it would be the closest-in-time U.S. patent art (June 2017, ~2 months before filing) and worth special scrutiny.

Part C — Bottom line on § 102 anticipation of the claims

• Claims 1 and 10 (the two independent claims) require a specific combination: (i) a ground truth built by annotation; (ii) a hit-map image identifying pixels that hit a lane marking; (iii) a first lane template from the previous view; (iv) generation of a fitted lane marking from the template + hit-map; and (v) training a confidence module using the ground truth plus the hit-map/fitted marking, where confidence is judged "using parameters of an arc of a circle fitted into the fitted lane marking." On the partial record I could retrieve, no single listed reference appears to disclose this combination; the older U.S. references are background/§ 103 material, and the likely-anticipatory candidates (the two CN publications) are unverified. A proper anticipation opinion requires the full documents.
• Dependent claims are where the cited art is more likely to land element-by-element: claim 14 (constraints: parallel markings; circular curvature; curvature < ~300 m; lane spacing ~3–4 m; edge-color difference) maps onto classic lane-modeling art such as the Crabtree/Banning/Wooten references; claims 2/12 (arcs-of-circles annotation) and 3/13 (train algorithm + generate hit-map) map onto the newer CN/CNN art. This is potential § 102/§ 103 relevance only — I have not verified any reference's disclosure.
• Claims 5–9 and 15–19 (confidence thresholding, accept/reject, failure-pattern determination, refinement with labeled data) appear distinctive; none of the retrieved references is a plausible anticipatory reference for them.

Part D — What I could NOT verify (explicit gaps)

1. The remainder of the (56) "U.S. Patent Documents … Continued" list on the '354 front page — several entries are cut off in the capture I obtained.
2. The full "Other Publications" list (the Bar Hillel survey appears to be one of possibly several NPL items).
3. The PCT International Search Report / Written Opinion for WO 2019/040670 A1 (the examiner-cited art there may differ from the U.S. face list).
4. The actual disclosure/titles of US 6,084,870, US 6,263,088, US 6,594,821, CN 106340197 A, and CN 106781591 A.
5. Whether US 9,672,446 is cited in '354 or only in a later family member.

Recommended retrieval to close these gaps: the '354 front-page PDF (all (56) entries), the file wrapper at USPTO PatentCenter/Global Dossier (form PTO-892 "List of References Cited by Examiner"), and the ISR for PCT/US2018/047608. I could not reach PatentCenter/PACER (paywalled) in this session.

Length/date note (consistency flag): As your earlier section already flagged, the task prompt states "Current Date: April 26, 2026," while the system date is 2026-09-29. The '354 patent family data and the litigation items in the prior summary are therefore reported as-of the sources' timestamps; nothing in this citation analysis depends on the 2026 date, since all listed references predate the 2017-08-22 filing.

If you can supply the complete front-page PDF or the PTO-892, I can finish the element-by-element § 102 chart (claim-by-claim, limitation-by-limitation) against each reference.

I'll research the prior art landscape for this patent before drafting the obviousness analysis.

Let me search for more specific prior-art references on the key claim elements.

Let me retrieve the examiner-cited references on the face of the patent and the corresponding family search reports.

Obviousness Analysis — U.S. Patent No. 10,816,354 under 35 U.S.C. § 103

Analytical caveat up front. I did not read the full text of every reference cited below; several were surfaced through search snippets, family search reports, and the patent's own "References Cited" fragment. I mark my confidence per reference. This is a technical/analyst assessment, not a legal opinion; invalidity must ultimately be proven by clear and convincing evidence. I also restate the date anomaly flagged in the earlier sections: the prompt says "April 26, 2026," the system date is 2026-09-29, and some retrieved sources post-date April 26, 2026.

A note on "the Prior Art section of this page." The Google Patents page for US 10,816,354 B2, as fetched, does not contain a conventional examiner-citation ("References Cited") list; its "Prior Art" section consists of the keywords ("lane; fitted; lane marking; module; hit"), the Cited By (7) block, and the Families Citing this family (22) block. Those two blocks are almost entirely forward citations that post-date the 2017-08-22 priority date and therefore are not § 102/§ 103 prior art by themselves. The usable prior art therefore comes from (i) the references actually named on the face of the patent as retrieved (US 6,084,870 A (Wooten et al., 2000); CN 106340197 A (1/2017); CN 106781591 A (5/2017); "Aharon Bar Hillel et al., Recent Progress in Road and Lane Detection — A survey"), plus (ii) references I retrieved that are date-qualified prior art to the '354 patent. I use both, labeled as such.

1. Effective filing date and the prior-art window

• '354 patent: filed 2017-08-22; priority 2017-08-22; claims are AIA-era (post-2013), so § 102(a)(1)–(a)(2) apply.
• Any reference published, or a U.S. patent/publication/WO publication effectively filed, before 2017-08-22 is available. Under § 102(a)(2), a U.S. patent granted after the critical date still qualifies if its effective filing date (provisional or non-provisional) precedes it.

All references below satisfy this window (verified or verifiable from the priority/filing data retrieved).

2. Level of ordinary skill in the art (PHOSITA)

A person having ordinary skill as of August 2017 would have a bachelor's or master's degree in electrical engineering, computer science, robotics, or a related field, and two to five years of experience in computer-vision-based lane/boundary detection, ADAS, or autonomous-driving perception, or equivalent. That person would be familiar with (a) supervised deep-learning segmentation/classification of road pixels, (b) geometric lane models (straight-line, parabolic, circular-arc, clothoid, spline), (c) model-fitting with RANSAC or least squares, (d) temporal filtering/tracking of lanes across frames (Kalman/particle filters), (e) inverse-perspective / bird's-eye ("top-down" or, in the patent's language, "god's view") transformation, and (f) multi-sensor fusion of camera with IMU/GPS/HD-map priors.

The '354 patent's own specification confirms this baseline is ordinary: it treats the lane constraints (parallel markings; circular curvature; curvature < ~300 m; 3–4 m lane spacing; differing edge color) as known road priors, not as inventions.

3. Claim 1 — element-by-element mapping

The claim (as issued) recites six elements. Mapping to prior art:

Conclusion on claim 1: every element is disclosed or rendered obvious by the combination of Baidu (primary) + Ford + Kim/Li + a temporal-lane-tracking reference (EP 3084681). No element requires anything beyond ordinary skill.

4. The primary § 103 combination

Combination A — Baidu + Ford + Kim/Li (+ EP 3084681)

Baidu (US 10,074,020 B2 / EP 3171292 A1 / KR 20170041168 A), priority CN201510574875 (2015-09-10), published 2017-05-24.
Teaches: (i) DNN-generated pixel confidence map (≈ claim 1's "hit-map image"), (ii) candidate lane line extraction, (iii) computing a lane-line confidence from pixel confidences, (iv) filtering/rejecting the lane line when confidence is below a preset threshold, (v) use of GPS positioning data, and (vi) splicing results from at least two consecutive images. It therefore directly addresses elements (a), (b), (c), and the confidence/threshold core of (f).

Ford (US 2017/0369057 A1; US 10,336,326 B2), provisional 62/354,583 filed 2016-06-24; non-provisional 2016-11-28.
Teaches: front-camera image → geometric transformation to a birds-eye view → deep neural network determines lane position; and (in US 10,336,326) segmenting the birds-eye image and running the network per segment. This supplies the "god's view" / top-down representation (claim 20's coordinate frame) and reinforces element (b).

Kim et al., "Robust Lane Detection Based On Convolutional Neural Network and Random Sample Consensus," Springer (2014).
Teaches: CNN lane-pixel detection followed by model fitting (RANSAC) to the detected lane points — i.e., fitting a geometric curve (including circular-arc models used in lane detection) to the detector output. Supplies the "arc of a circle fitted into the fitted lane marking."

EP 3084681 A1 / WO 2015/150256 A1 (published 2016-10-26).
Teaches: tracking detected lane markings across frames with a recursive state estimator, estimating separate progressions for tracked lane markings and associating offsets to them. Supplies element (d) (the previous-view "lane template") and the prediction half of element (e).

Motivation to combine (KSR / MPEP 2143):

1. Same field, same problem. All four references are in automotive lane/boundary detection for ADAS or autonomous driving and address the same problem the '354 patent recites as its own: robustly localizing lane markings from a moving vehicle.
2. Predictable, complementary techniques. Baidu already uses a deep network to produce a per-pixel lane map and to score a fitted lane line; Ford already uses a deep network on a bird's-eye view; Kim already shows CNN detection feeding a geometric/RANSAC curve fit; EP 3084681 already shows frame-to-frame lane tracking. Combining a detector, a geometric fitter, a tracker, and a confidence threshold is the routine architecture of the field — a "combination of familiar elements according to known methods" that yields no more than the expected result (KSR Int'l v. Teleflex).
3. Design need / finite solutions. A PHOSITA seeking a real-time, self-diagnosing lane detector would be motivated to (i) smooth noisy frame-by-frame detections using a prior-frame template (reduces jitter, improves speed — exactly the benefit the '354 specification attributes to the lane template), and (ii) use a low-parameter geometric lane model (circle arc/parabola) because highway curvature is low and the model is compact and robust — the '354 specification itself lists this as a known road prior.
4. Reasonable expectation of success. Each sub-combination was individually known to work in real vehicle systems; there was no technological barrier to integrating them.

Combination B — Li et al. + Ford + Baidu + EP 3084681 (alternative primary)

Jun Li et al., "Deep Neural Network for Structural Prediction and Lane Detection in Traffic Scene," IEEE Trans. Neural Networks & Learning Systems 28(3):690–703 (published 2016-02-16) was cited as an "X" (novelty-destroying) reference against claims 1–4 and 6–15 in the search report of a different application (EP 3441909 A1). That is strong evidence that this single printed publication was considered to disclose the core deep-learning structural lane-detection pipeline. Combined with Ford (bird's-eye), Baidu (pixel confidence + GPS + threshold filtering), and EP 3084681 (temporal tracking), it independently renders claim 1 obvious. I flag that the Jun Li X-citation belongs to a different patent family (EP 3441909 A1), so it is evidence of the state of the art, not a direct statement about the '354 prosecution.

5. Dependent claims

Claim 2 / 12 — "annotating the … lane markings using arcs of circles." Obvious over Kim et al. (curve fitting) and Bar Hillel (2014) (lane-model survey) — annotating a lane by fitting a circular arc is a direct application of the known lane geometry the specification itself calls a road prior.

Claim 3 / 13 — "training the lane detection algorithm … ; generating another hit-map image." Baidu (trains the deep network on marked samples; runs it to produce pixel confidences) and Li et al. (structural prediction). This is the ordinary supervised-learning loop.

Claim 4 / 14 — "adjusting the first lane template … according to one or more rules," plus constraints: (i) markings parallel, (ii) curvatures circular, (iii) curvature < ~300 m, (iv) lane spacing ~3–4 m, (v) edge color differs from non-marking road. Each constraint is a textbook physical prior of highway geometry. The Bar Hillel survey (2014) — cited on the face of the patent — catalogues these lane models; lane width ≈ 3.5–3.75 m and low highway curvature are standard civil-engineering/map facts (and the '354 specification concedes they are "priors or constraints"). Using them to regularize/constrain a fit is the routine application of known engineering rules.

Claim 5 / 15 — "confidence level … above or equal to a threshold … ; outputting … as a predicted lane marking." Baidu expressly discloses preset-threshold accept/reject of candidate lane lines by confidence. Near-verbatim anticipation of this limitation in substance.

Claim 6 / 16 and 8 / 18 — comparing the accepted (or rejected) fitted lane marking to ground truth → "failure pattern." Comparing a detector output against annotated ground truth and identifying the failure mode is the definition of routine model validation; Baidu, Li et al., and Huval et al. all train/validate against annotated data. The only added wrinkle — mining misclassifications for a "failure pattern" — is conventional error analysis that a PHOSITA would apply when improving a model.

Claim 7 / 17 — "confidence level … below a threshold … ; rejecting the fitted lane marking." Baidu discloses rejecting candidate lane lines below a confidence threshold.

Claim 9 / 19 — "refining the confidence module by adding labeled data." Iterative/active re-training with additional labeled data is routine ML practice and is taught generically by Huval et al. and Li et al.; no unexpected result is claimed.

Claim 11 — "generating a second lane template associated with the current view … using the parameters; using [it] for … a next view." This is the temporal feedback loop of EP 3084681 (lane state carried from frame N to frame N+1) and Baidu (splicing from ≥2 consecutive images). Obvious.

Claim 20 — the "god's view" coordinate frame (vehicle at origin; y-axis fore–aft; x-axis perpendicular). Ford (US 10,336,326 / US 2017/0369057) discloses the bird's-eye view transformation of the camera scene; defining a vehicle-centric orthogonal frame (y = heading, x = lateral) is the standard robotics convention and carries no patentable weight. Also note this limitation is definitional, not structural.

6. Why a PHOSITA would have been motivated to combine (consolidated)

Under KSR and MPEP 2143, the motivation is strong because:

1. Identical field of endeavor — camera-based lane/boundary detection for vehicles (all references).
2. Common problem and known interaction — the references address the same "detect lane markings robustly in real time" problem, and the elements (pixel classifier → geometric fit → temporal template → confidence gate → threshold accept/reject) are known to interact in a predictable, additive way.
3. Obvious-to-try / finite solutions — using a low-degree geometric lane model (arc/parabola), a prior-frame template, and a confidence threshold were among a small number of well-known options for stabilizing a noisy detector.
4. Design incentives explicit in the references themselves — Baidu's stated goal is efficient, accurate lane recognition with confidence-based filtering; Ford's is to reuse a trained network on a bird's-eye image; EP 3084681's is to track lanes across frames. Combining them yields the predictable aggregate benefit the '354 specification also states (better precision, faster fitting, self-diagnosis).
5. No teaching away, no unexpected result — none of the references criticizes the others' approaches, and the '354 patent claims no criticality or unexpected property for the "arc of a circle" parameterization or for training the confidence module on the hit-map/fitted marking.

7. Counter-considerations a careful analyst should preserve

• The specific "confidence module trained on the ground truth and the hit-map/fitted marking" is the element I consider least squarely disclosed by any single reference. Baidu computes lane confidence analytically (mean pixel confidence × quantity ratio) rather than by training a separate verifier network. However, a learned verification/scoring head trained on labeled detections is a well-known ML technique, and Li et al./Huval et al. provide the labeled-training framework — so this is a § 103 (obvious design choice) point, not a § 102 (anticipation) point. I did not identify a single reference that anticipates claim 1 in its entirety; my conclusion is one of obviousness, not anticipation.
• I could not retrieve the complete "References Cited" list on the face of US 10,816,354, nor the full EP 3673340 / CN 111052132 B search reports. Those documents may contain examiner-applied X/Y art that is even more on-point than what I assembled. The strongest known X-type art for the core deep-learning pipeline (Jun Li et al.) came from a sibling-but-different application's search report (EP 3441909 A1), not from the '354 family — a genuine gap.
• Dates. Verify each reference's effective date precisely: Ford's provisional (2016-06-24) precedes 2017-08-22, but its publication (2017-12-28) does not; it is prior art only via § 102(a)(2) "effectively filed" status. Baidu's EP publication (2017-05-24) is prior art directly; its US grant (US 10,074,020 B2) qualifies via its CN priority (2015-09-10).
• Assignee/litigation posture. As the earlier sections found, no litigation asserting the '354 patent was confirmed, so there is no judicially developed claim construction or invalidity record to leverage; and the reported asserted patents in CreateAI, Inc. v. Bot Auto, Inc., No. 4:25-cv-03371 (S.D. Tex.) were US 11,577,783 / US 12,190,609 (and a '783-type reference), not the '354 patent. Any § 103 challenge would proceed on the art alone.
• Secondary considerations (commercial success, licensing, copying, long-felt need) are not documented on the available record, but note the technology's adoption in the autonomous-truck industry could be argued by the patentee as evidence of non-obviousness; no nexus evidence was found.

8. Bottom line

On the retrieved record, claim 1 is vulnerable to a § 103 obviousness challenge built on Baidu (US 10,074,020 B2 / EP 3171292 A1) as the primary reference for the per-pixel lane map, the lane-line confidence computation, GPS use, and threshold-based accept/reject; in view of Ford (US 10,336,326 B2 / US 2017/0369057 A1) for the bird's-eye ("god's view") deep-network lane detection; Kim et al. (2014) or Li et al. (2016) for CNN-based detection with geometric/curve (arc) fitting; and EP 3084681 A1 / WO 2015/150256 A1 for the previous-view lane template/tracking. Dependent claims 2–9, 11–20 map to the same references plus the well-known road priors catalogued in the on-face Bar Hillel (2014) survey and the on-face references US 6,084,870 A, CN 106340197 A, and CN 106781591 A.

The principal residual non-obviousness argument available to the patentee is the trained confidence module (as opposed to an analytically computed lane-line confidence) and the specific "arc of a circle" parameterization of the confidence check; both are, in my assessment, design choices within the ordinary skill level, but they are the points where a § 103 rejection would most likely be contested and where a full-text review of the true examiner-cited art is needed to reach a firm conclusion.