US 10574894

I'll verify the patent details and litigation posture with searches, then summarize based on the authoritative patent text provided.

Let me verify the litigation/PTAB posture and the claim set.

Let me try to retrieve the exact independent claim language.

US Patent 10,574,894 — Verified Summary

Important caveat up front: I was able to verify every bibliographic field below against the authoritative full text you supplied (Google Patents, fetched 2026‑09‑24) plus corroborating public sources. However, I could not retrieve the verbatim claim text in this session — the search only returned secondary quotations of claim limitations. The per‑claim overview below is therefore reconstructed from the specification, the abstract, and third‑party quotations of claim 1/claim 11 language. Where I am inferring, I say so.

---

1. Bibliographic data (verified)

Priority chain (recorded on the face of the Google Patents page): US 16/548,453 is itself a continuation‑type filing with a priority date of 2018‑05‑18. Later GoPro continuations claiming priority through this case include US 11,025,824 (16/787,693), US 11,363,197 (17/334,150), US 11,696,027 (17/837,013), US 12,256,147 (18/345,229), and US 2025/0220306 A1 (19/082,025).

---

2. Abstract (verbatim)

"Images with an optical field of view are captured by an image capture device. An observed trajectory of the image capture device reflects the positions of the image capture device at different moments may be determined. A capture trajectory of the image capture device reflects virtual positions of the image capture device from which video content may be generated. The capture trajectory is determined based on a subsequent portion of the observed trajectory such that a portion of the capture trajectory corresponding to a portion of the observed trajectory is determined based on a subsequent portion of the observed trajectory. Orientations of punch-outs for the images are determined based on the capture trajectory. Video content is generated based on visual content of the images within the punch-outs."

---

3. Plain‑language overview of the independent claims

The disclosure's core idea is "look‑ahead" stabilization: instead of guessing where the camera is going using only past motion, the system examines future camera motion to decide whether a given wobble was intentional (e.g., a deliberate pan) or unintended (e.g., a bump or hand shake), then builds a smoother "virtual camera" path and crops ("punches out") each frame accordingly.

Claim 1 — method claim (independent). Plain language: You capture wide‑field images with a camera that has a larger optical field of view than the field of view you ultimately want to show. You also record camera position over the capture period (from a gyroscope/accelerometer/IMU, not from the images). From that position data you build an "observed trajectory." You then build a capture trajectory — a virtual camera path — where a portion of the path for one moment is derived from a later (subsequent) portion of the observed trajectory, so the path is smoother/less jittery than the observed path. Per the parties' filings, claim 1 also requires that the capture trajectory "include a path that minimizes a combination of rotational velocity and rotational acceleration of the housing, the combination of the rotational velocity and the rotational acceleration weighing them equally or differently." Then you orient the punch‑out (capture field of view) within each wide image based on the capture trajectory, and generate the video from the pixels inside those punch‑outs.

Claim 11 — system/apparatus claim (independent). Plain language: The same pipeline, but expressed as an apparatus — one or more processors (and, per the specification, image sensor/position sensor carried by a camera housing) configured to determine the observed trajectory, determine the look‑ahead capture trajectory that minimizes the weighted combination of rotational velocity and rotational acceleration, determine the punch‑out orientations, and generate the stabilized video. The IPR record treats claim 11 as the independent system counterpart to claim 1, containing the same key minimization/weighting limitation.

Dependent claims (2–10 from claim 1; 12–20 from claim 11). From the specification and the litigation record, the dependent claims add limitations such as: the smoother trajectory being characterized by less jitter; the minimization being performed subject to a set of constraints; and specific constraints — a margin constraint derived from the difference between optical FOV and capture FOV, a trajectory constraint derived from a subsequent portion of the observed trajectory, and a target constraint derived from the position of a target (e.g., a person) in the images. (I flag these as inferred groupings; I did not obtain the verbatim dependent‑claim text, so the exact claim‑to‑limitation mapping should be confirmed against the granted claims at USPTO PatentCenter.)

---

4. Litigation and PTAB posture (verified, and relevant to the claim scope)

• IPR2024-01186 (PTAB). Filed 2024‑07‑18 against the '894 patent; institution decision 2025‑01‑22; Final Written Decision issued 2025‑12‑10 determining no challenged claims unpatentable; the case is now marked "Final Written Decision – Appealed." Sources disagree on the petitioner: an independent PTAB tracker identifies the petitioner as Arashi Vision Inc. d/b/a Insta360, while the Google Patents litigation block lists the proceeding without naming a petitioner (the "Petitioner:" field is blank in the supplied text). Treat the petitioner identity as not fully confirmed from the material supplied. Note: some search results discussing a Final Written Decision finding claims unpatentable relate to a different patent (US 8,014,284, Unified Patents) and must not be conflated with this case.
• Court of Appeals for the Federal Circuit, case 26‑1428 — listed on the Google Patents page under "Family has litigation." I could not retrieve the docket text itself; the docket number is consistent with an appeal from the 2024‑01186 Final Written Decision, but I cannot confirm the parties or issues from an authoritative source. Flagging as unverified.
• ITC Investigation No. 337‑TA‑1400 (Certain Cameras, Camera Systems, and Accessories Used Therewith), brought by GoPro against Arashi Vision/Insta360. The '894 was one of five asserted utility patents. GoPro's asserted claims of the '894 (claims 2–4 and 6–20) were terminated from the investigation on the complainant's own motion (ALJ initial determination No. 24; Commission declined review 2025‑01‑31), leaving claims 1 and 5. In the final determination (announced ~2026‑02‑26), the Commission found no violation as to all five utility patents, including a specific finding that the accused products do not infringe the '894 patent. Only the design patent D789,435 supported a limited exclusion order directed at legacy Insta360 Ace models.
• District Court, C.D. Cal. No. 8:24‑cv‑00681 — a related district‑court action is listed on the Google Patents page. I did not verify its docket contents.

---

5. Uncertainty / verification notes

1. Claim text is not verbatim. Only the abstract and description are authoritatively before me from your supplied document; the abstract of the independent claims above is reconstructed. The one claim‑language quotation I found ("…minimizes a combination of rotational velocity and rotational acceleration of the housing, the combination of the rotational velocity and the rotational acceleration weighing them equally or differently") comes from a Patent Owner expert declaration in the IPR, not from the granted claims themselves.
2. Which claims are independent. The IPR record refers only to "independent claims 1 and 11." I could not confirm whether any of claims 12–20 is separately independent, or whether all of them depend from claim 11.
3. Petitioner identity in IPR2024‑01186. The Google Patents record shows a blank petitioner field; the Insta360 attribution comes from a third‑party tracker.
4. CAFC 26‑1428. Confirmed only as a listing; the parties, appeal type, and issues are unverified.
5. I did not auto‑correct any identifiers. US 10,574,894 ("'894") is distinct from US 10,432,864, US 11,025,824, US 11,353,489, US 11,979,662, US 12,256,147, EP 3,854,070, and US 8,014,284 — several of which appear in search results for similar subject matter and should not be merged with this patent.

I'll search for litigation involving US patent 10574894, drawing on litigation databases and the patent record itself.

Let me gather details on the Federal Circuit appeal and the district court case.

Let me confirm details of the Federal Circuit appeal and the IPR outcome.

Litigation Involving U.S. Patent No. 10,574,894

Patent identified: U.S. Patent No. 10,574,894 B2 ("Systems and methods for stabilizing videos"), Application No. 16/548,453, filed 2019-08-22, granted 2020-02-25, priority date 2018-05-18, assignee GoPro, Inc. Inventors: Thomas Derbanne, César Douady, Maxim Karpushin.

The '894 patent has been asserted in three related U.S. proceedings (an ITC Section 337 investigation, a parallel district court action, and a PTAB inter partes review), plus an appeal. It has also been listed in the Google Patents "Family has litigation" data as involving the Court of Appeals for the Federal Circuit. Details below.

⚠️ Verification note: Litigation statuses move quickly; the following reflects the best available information as of the search date. Case-status characterizations from secondary aggregators (Unified Patents, Darts-IP, IPRdaily, Chinese-language press) should be confirmed against the primary dockets (PACER, ITC EDIS, PTAB, CAFC) before being relied on.

---

1. ITC Section 337 Investigation — Certain Cameras, Camera Systems, and Accessories Used Therewith

Procedural history for the '894 patent specifically:

• The Commission terminated claims 2–4 and 6–20 of the '894 patent based on Complainant's own motion (ALJ Order No. 24 of Jan. 13, 2025, not reviewed by the Commission; see notice published Feb. 5, 2025). This left claims 1 and 5 at issue.
• Final Initial Determination (July 10, 2025): ALJ Hines found no violation for all five utility patents, including the '894 patent. The ALJ recommended a limited exclusion order and cease-and-desist order only as to the D789,435 design patent (with a 0% bond based on price-differential analysis).
• Both sides petitioned for review (July 25, 2025). GoPro sought review of the no-violation findings on all five utility patents; Insta360 challenged the design-patent infringement finding. On September 11, 2025, the Commission determined to review only the '840 patent, the '052 patent, and the D'435 patent — and declined to review the no-violation findings for the '894, '832, and '413 patents, i.e., the '894 win for Insta360 became final at the Commission level.
• Commission Final Determination (February 26, 2026): Investigation terminated in its entirety. GoPro's six patent allegations all resolved in Insta360's favor on the utility patents; the only infringement was a previous-generation Insta360 Ace-series product under the D789,435 design patent (that model already ceased U.S. exports; the new design was found non-infringing). Limited exclusion order and cease-and-desist order issued only as to the design patent.
• Source: ITC Final ID (ALJ Hines, July 10, 2025) and Commission Notice, 91 Fed. Reg. (Mar. 3, 2026); Morrison Foerster client release ("complete victory for Insta360," Feb. 26, 2026).

GoPro's appeal rights: GoPro had 60 days from entry of the final determination to appeal to the U.S. Court of Appeals for the Federal Circuit. Whether GoPro filed a notice of appeal on the utility patents (including '894) is not confirmed in the sources retrieved.

---

2. Parallel District Court Action — GoPro, Inc. v. Arashi Vision Inc. et al.

Sources: GoPro complaint (8:24-cv-00681) excerpt; Justia docket listing; IPRdaily and Insta360 (影石创新) announcements of Feb. 27–28, 2026.

---

3. PTAB Inter Partes Review — Arashi Vision Inc. (d/b/a Insta360) v. GoPro, Inc.

Sources: PTAB docket data (ipverse/GreyB and Docket Alarm); PTAB Final Written Decision (Dec. 10, 2025) "Determining No Challenged Claims Unpatentable."

---

4. Court of Appeals for the Federal Circuit Appeal

This is the "US case filed in Court of Appeals for the Federal Circuit" flagged in the Google Patents litigation data for the '894 patent family. Source: Google Patents family-litigation data linking to Unified Patents (CAFC case 26-1428); PTAB docket showing the Feb. 6, 2026 Notice of Appeal.

---

5. Related / Not Directly On The '894 Patent

• GoPro's possible ITC appeal: GoPro retains the right to appeal the ITC's final determination (including the '894 no-violation finding) within 60 days; no such appeal has been confirmed in the retrieved sources.
• Insta360's Chinese mirror litigation: Insta360 (and its subsidiary 深圳前海影石) filed patent-infringement suits in China (Jiangsu High Court; Shenzhen Intermediate Court; Changsha Intermediate Court) against GoPro and its Chinese entities/distributors (2024). These involve Insta360's own patents, not U.S. 10,574,894, and are noted only as context for the broader GoPro–Insta360 dispute.

---

Summary Table

---

Caveats on precision

• The District Court filing date is inferred as on/about March 29, 2024 (contemporaneous with the ITC complaint) — I did not independently confirm the exact CACD filing date from a primary docket; the docket number 8:24-cv-00681 and its listed title/parties are corroborated by the Justia docket listing and the complaint document.
• The exact filing date of the CAFC appeal (26-1428) was not independently verified beyond the Feb. 6, 2026 PTAB Notice of Appeal; the CAFC docket number itself comes from Google Patents/Unified Patents family-litigation data.
• I found no other litigation (e.g., additional district court suits or other ITC investigations) asserting U.S. 10,574,894 beyond those listed above.

I'll research the PTAB proceedings for this patent, starting with the IPR referenced in the patent metadata.

I found the key proceeding. Let me gather the claim-level details and verify whether any other IPRs exist on this patent.

PTAB Activity on US 10,574,894 — Analyst Report

⚠️ Data-quality flag up front: The structured "PTAB proceedings on file" block states that the USPTO ODP API returns no AIA trial proceedings for this patent as of the most recent ingest. That is incorrect (or at least incomplete). Web search surfaces one AIA trial that ran to a Final Written Decision and is now on appeal: IPR2024-01186. I have included it below and flagged it as not sourced from the ODP block.

Second flag — petitioner identity: The Google Patents "Family has litigation" panel for this patent renders a line reading PTAB case IPR2024-01186 filed (Final Written Decision) … Petitioner: … "Unified Patents PTAB Data" by Unified Patents is licensed under a CC BY 4.0 License. That "Unified Patents" string is the dataset attribution for Google's PTAB feed, not the petitioner. The actual petitioner is Arashi Vision Inc. (d/b/a Insta360) and its U.S. subsidiary. Do not brief this case as a Unified Patents challenge.

---

Proceedings overview

Total AIA trial proceedings on US 10,574,894: one (1) — IPR2024-01186. Breakdown: 0 proceedings with claims canceled, 1 proceeding with all challenged claims sustained, 0 settled, 0 institution-denied. Bottom line for a defendant: this patent survived its only IPR intact — the Board issued a Final Written Decision Determining No Challenged Claims Unpatentable under 35 U.S.C. § 318(a) on 2025-12-10 — so it is not a patent that a canned obviousness IPR will knock out, and any invalidity theory has to beat a Board finding that Kwatra does not teach the claimed "minimize rotational velocity and rotational acceleration" limitation. The one piece of genuinely good news for defendants is that GoPro lost on infringement of this patent at the ITC (Insta360's accused products were found non-infringing as to the '894 patent), meaning GoPro has so far failed to prove that anyone actually practices these claims.

---

IPR2024-01186 — Arashi Vision Inc. (d/b/a Insta360) v. GoPro, Inc.

(Only proceeding. Most impactful by default. Claims sustained, not invalidated.)

• Type: Inter Partes Review (35 U.S.C. §§ 311–319)
• Filed: 2024-07-18
• Status (verbatim from docket data): "Final Written Decision - Appealed." Plain-English gloss: the Board has issued its Final Written Decision, and Petitioner has noticed an appeal to the Federal Circuit; the PTAB trial itself is terminated (2025-12-10).
• Judge panel (APJs): William V. Saindon, Nabeel U. Khan, and Brian D. Range. (Tech Center 2600; Art Unit 2698.)
• Petition grounds: Obviousness under 35 U.S.C. § 103. Petitioner challenged claims 1–20 (the full claim set of the '894 patent, per the ITC's identification of "claims 1–20 of [U.S. Patent No. 10,574,894]"). Ground 1 was asserted as "Zhou and Kwatra render claims 1–20 obvious." A further combination adding Shi was asserted against claims 2, 3, 7, 12, 13, and 17 for the "intentional motion preservation" aspect. Petitioner's exhibit list included U.S. Patent Nos. 9,674,438; 8,531,535; 10,462,370; and 9,787,902, plus a declaration of Dr. Aggelos Katsaggelos. The dispositive dispute centered on whether Kwatra taught or suggested minimizing rotational velocity and rotational acceleration. (Ground detail is drawn from third-party case summaries of the as-filed petition and from the parties' oral-argument briefing; treat the specific § 103 ground numbering with mild caution — the FWD itself is the authoritative artifact.)
• Institution decision: Instituted (grant) on 2025-01-22. The Board granted review, finding a reasonable likelihood of prevailing at least as to the challenged claims. (The panel's full reasoning in the institution decision is not reproduced in the sources I could retrieve; I will not fabricate it. Note that the panel did issue a preliminary claim construction in the companion '840 proceedings, not the '894 case.)
• Final Written Decision: Issued 2025-12-10, captioned "Final Written Decision Determining No Challenged Claims Unpatentable 35 U.S.C. § 318(a)." That caption is the operative disposition: the Board held that Petitioner failed to prove any challenged claim unpatentable, so all challenged claims of the '894 patent were sustained. On the merits, the Board credited GoPro's position that Kwatra's cost function applied only to camera-path translational components (x, y, z) and did not teach minimizing rotation (Kwatra expressly avoided rotation due to computational complexity), so the combination did not disclose the claim-1 requirement of a path minimizing a combination of rotational velocity and rotational acceleration. Caveat on granularity: I could not retrieve the FWD body text, so I cannot quote the panel verbatim or confirm the claim-by-claim disposition line-by-line. What is confirmable is the caption-level verdict — no challenged claims unpatentable — which necessarily means claims 1–20 (the challenged set) all survived. I am not representing that the Board separately held any individual dependent claim patentable in so many words; the FWD's disposition is global.
• Settlement / termination: No settlement. The proceeding ran to a merits FWD and terminated on the FWD date (2025-12-10). There was a motion-to-seal practice (sealed and public versions of PO demonstratives, sur-replies, and exhibit 2052), and a Patent Owner motion to strike (2025-08-26) with Petitioner's opposition (2025-09-03) — i.e., a contested record, not a walk-away.
• Appeal: Yes. Petitioner Arashi Vision (U.S.) LLC (d/b/a Insta360) filed a Notice of Appeal on 2026-02-06 (Paper 51). The appeal is docketed at the U.S. Court of Appeals for the Federal Circuit as No. 26-1428 (Arashi Vision US LLC v. GoPro Inc), docketed 2026-02-11, appeal from IPR2024-01186. As of the most recent data retrieved, the appeal outcome is pending and no merits panel has been assigned. (Confirm current status on the CAFC docket / CourtListener before relying on it — docket data ages quickly.)
• Defensive value: This is a pro-patent-owner outcome, so an IPR-based invalidity defense is harder here, not easier — the Board has already rejected the Zhou/Kwatra § 103 theory on the one limitation that matters (rotation-minimization), and that reasoning will carry weight with the Board and with district judges under Phillips-style review. But the FWD also hands a defendant a useful roadmap: it narrows Kwatra out of the "minimize rotational velocity and acceleration" space, which tells you exactly which element any new prior-art attack must attack, and with what kind of reference. Because the FWD is now appealed, the affirmance/vacatur risk cuts both ways — a Federal Circuit vacatur would reopen the invalidity door; affirmance would harden the patent.

---

Strategic summary

Claim status on US 10,574,894. The patent has one IPR, and in it no claim was canceled. The challenged set was claims 1–20, and the FWD determined no challenged claims unpatentable, so every challenged claim is, for now, SUSTAINED. There are currently no CANCELED claims and no UNTESTED claims on this record (the petition reached the full claim set). Two nuances matter: (i) because the FWD disposition is a global "no challenged claims unpatentable" rather than a claim-by-claim opinion I could verify, I do not assert which individual dependent claims were separately discussed; and (ii) the FWD is on appeal at the Federal Circuit (No. 26-1428), so "sustained" is subject to appellate review. If GoPro's demand letter relies on claims 1–20, the troll/prisoner's-dilemma framing does not apply — none of those claims is dead, and you cannot win an early motion on IPR estoppel/collateral estoppel against GoPro.

Estoppel landscape. The FWD triggers § 315(e)(2) estoppel — but it runs against the Petitioner (Arashi Vision/Insta360) and its real parties-in-interest and privies only, in the C.D. Cal. action (No. 8:24-cv-00681) and the ITC investigation, as to any ground Insta360 raised or reasonably could have raised in IPR2024-01186 (i.e., the Zhou/Kwatra/Shi § 103 combinations). For a different, non-privy defendant, § 315(e)(2) imposes no estoppel at all — you remain free to raise the very same Zhou/Kwatra ground, though as a practical matter you would be relitigating a theory the Board just rejected. Conversely, because the case went against the petitioner, there is no § 315(e)(1)/§ 315(e)(2) estoppel benefit flowing to defendants from a canceled claim. The realistic value of IPR2024-01186 to a new defendant is evidentiary and strategic, not estoppel-based: it is a fully developed record on how the Board reads the "minimize rotational velocity and rotational acceleration" limitation, and a demonstration that Kwatra (as presented) is insufficient.

Pattern signals. (1) Same petitioner, multiple IPRs, different GoPro patents. Insta360 filed a coordinated family — IPR2024-01164 and IPR2024-01165, both against U.S. Patent No. 10,958,840 (a sibling video-stabilization patent), plus IPR2024-01186 against the '894 patent. All three were argued at a single consolidated oral hearing (transcript public version 2025-11-04). Insta360 also filed IPR2024-01434 against GoPro design patent D789,435, which was denied institution. (2) No defensive aggregator. Despite the misleading "Unified Patents" string in Google's litigation panel, there is no Unified Patents or other defensive-aggregator IPR on this patent; the petitioner is a commercial competitor. (3) Patent owner litigates aggressively on multiple fronts. GoPro asserted the '894 patent twice in parallel — ITC Inv. No. 337-TA-1400 and C.D. Cal. No. 8:24-cv-00681 (stayed pending the ITC) — and the ITC's Final Initial Determination (2025-07-10) found no infringement as to the '894 patent, which the Commission's final determination affirmed on 2026-02-26 (no violation as to the '894 patent). So while the patentability battle went to GoPro, the infringement battle went to Insta360. (GoPro had a 60-day window from the 2026-02-26 ITC final determination to appeal to the Federal Circuit; I have no reliable information on whether it did, and I won't guess.)

---

Recommended next steps

1. Do not treat the '894 patent as IPR-weakened. The only FWD reads "Determining No Challenged Claims Unpatentable 35 U.S.C. § 318(a)" (2025-12-10). If a demand letter asserts claims 1–20, every asserted claim is currently alive. Any early-disposition strategy must be built on infringement or on new art, not on the premise that IPR canceled the claims.
2. Pull the actual FWD and the institution decision from PTAB E2E (IPR2024-01186) before drafting anything — I confirmed the caption-level result but could not verify the body text or a claim-by-claim breakdown. Verify before quoting. The panel is Saindon, Khan, and Range; Tech Center 2600 / Art Unit 2698.
3. Track the appeal. Federal Circuit No. 26-1428 (Arashi Vision US LLC v. GoPro Inc), notice of appeal 2026-02-06, CAFC-docketed 2026-02-11, appeal from IPR2024-01186. Outcome pending, no merits panel assigned as of the last data retrieved. A vacatur reopens the invalidity attack; affirmance locks in Kwatra's insufficiency. Monitor via the CAFC docket and CourtListener; the PTAB appeal number may also appear under the Board's "Appeals" tab.
4. Exploit the ITC infringement loss. For any defendant facing assertion of the '894 patent, the Commission's 2026-02-26 final determination (affirming the 2025-07-10 Final Initial Determination, ALJ Doris Johnson Hines) finding Insta360's accused products non-infringing as to the '894 patent is powerful non-infringement evidence for products in the same accused-product family — obtain the ITC public record (Inv. No. 337-TA-1400) for the non-infringement findings and stipulations on representative accused products.
5. If you plan your own IPR on the '894 patent, remember that the § 315(e)(2) bar does not bind you (you are not a privy of Insta360), but the Board's rejection of the Zhou/Kwatra rotation-minimization theory is now on the record — a new petition should target claim 1's rotation-minimization limitation with art Kwatra does not supply, and be prepared for the Board to follow its own reasoning. There is no active proceeding with live trial deadlines on this patent (the trial terminated 2025-12-10), so there is no institution-decision/one-year statutory clock currently running on the '894 patent.
6. Re-run the ODP check. The structured "no PTAB activity" result is wrong on this patent. Before any filing that depends on the PTAB record, verify directly against USPTO PTAB E2E / Patent Trial and Appeal Board End-to-End system for IPR2024-01186 and confirm whether any follow-on petition (e.g., a second IPR by a different petitioner) has been filed since the last ingest.

Authoritative links to verify (do not cite until opened):

• IPR2024-01186 (PTAB E2E / Docket Alarm): https://www.docketalarm.com/cases/PTAB/IPR2024-01186/Arashi_Vision_Inc._%28d-b-a_Insta360%29_v._GoPro_Inc/
• FWD caption "Determining No Challenged Claims Unpatentable," 2025-12-10: PTAB E2E, IPR2024-01186, Paper 50 (per the case-documents listing)
• CAFC No. 26-1428: https://ai-lab.exparte.com/case/cafc/26-1428/arashi-vision-us-llc-v-gopro-inc (verify against PACER/CourtListener)
• ITC Inv. No. 337-TA-1400, Final ID 2025-07-10 and Commission final determination 2026-02-26

Reliability note: the proceeding number, parties, filing date, institution date, judge panel, FWD date and caption, appeal date, and CAFC docket number are corroborated across multiple independent sources and I state them with confidence. The petition's specific ground numbering and the FWD's internal claim-by-claim reasoning are sourced from third-party summaries and the parties' briefing, not from the FWD text itself, and are flagged accordingly — verify against the primary documents before relying on them.

I'll reconstruct the assignment chain. Let me first pull the recorded assignment data and cross-reference the litigation/ownership context.

Let me pull the specific reel/frame records and the GoPro 2025 financing context.

US 10574894 — Ownership & Assignment Reconstruction

Data-coverage note (read first): The authoritative patent text you supplied carries Google Patents' legal events feed, which lists conveyance type, party names, and event dates, but the fetched page does not expose USPTO reel/frame numbers or the correspondent of record for the recordings on this specific patent. I was unable to pull the underlying Assignment Center reel/frame entries or correspondent names within this session, so those cells below are marked [not retrieved] rather than guessed. The event list itself (types, assignors, assignees, dates) is reliable and comes from the patent's own legal-events record.

---

Inventors

The three inventors are the GoPro video‑stabilization group (the "HyperSmooth" pipeline team, which GoPro staffs largely out of France). Each conveyed to GoPro on the filing date — a normal employee/inventor-to-company assignment, not a fire‑sale pattern.

Unusual-pattern check: No indication any inventor departed within 12 months of filing. There is no recorded second assignment or employer change for any of the three, so the "all inventors leave → portfolio sale" tell is not present.

---

Original assignee

GoPro, Inc. (Delaware corporation; NASDAQ: GPRO), 3000–3025 Clearview Way, San Mateo, CA.

• Product embodying the claims: Yes. The patent (a look‑ahead trajectory‑smoothing stabilizer) is core to GoPro's in‑camera "HyperSmooth" stabilization shipped in HERO‑series action cameras.
• Primary line of business: Action/360 cameras, mounts, and consumer imaging accessories.
• Current status: Operating, but financially stretched. On 2025‑08‑04 GoPro raised a $50M second‑lien term loan (lender Mateo Financing, LLC; Farallon Capital Management, L.L.C. as administrative/collateral agent) and granted Wells Fargo Bank, N.A. a first‑priority lien on all IP under its amended revolver. Not in bankruptcy; no Chapter 7/11 filing surfaced.

---

Assignment timeline

Records retrieved from the patent's legal‑events feed (Google Patents mirror of USPTO records). Order is chronological; reel/frame and correspondent could not be retrieved for this patent and are flagged.

• 2019‑08‑22 (recorded) / execution date [not retrieved] — Reel [not retrieved]

  • Conveyance: Assignment (assignment of assignors' interest)
  • Assignor: Karpushin, Maxim; Douady, César; Derbanne, Thomas
  • Assignee: GoPro, Inc.
  • Correspondent: [not retrieved]
  • Context: Initial inventor‑to‑company assignment taken on the filing date (no reorg, no transfer to a third party).

• 2020‑10‑19 (recorded) — Reel [not retrieved]

  • Conveyance: Security Interest / Security Agreement (grant of security interest)
  • Assignor: GoPro, Inc.
  • Assignee: [JPMorgan Chase Bank, N.A., as Administrative Agent](/asserters/jpmorgan-chase-bank-n-a-as-administrative-agent)
  • Correspondent: [not retrieved]
  • Context: Securitization — GoPro pledged its patent portfolio as collateral under its JPMorgan‑agented credit facility. Ownership did not change.

• 2021‑01‑25 (recorded) — Reel [not retrieved]

  • Conveyance: Release (release of patent security interest)
  • Assignor: JPMorgan Chase Bank, N.A., as Administrative Agent
  • Assignee: GoPro, Inc.
  • Correspondent: [not retrieved]
  • Context: Lien release on payoff/refinancing — JPMorgan exited; GoPro moved its secured facility to Wells Fargo (Wells Fargo revolver dated Jan 2021). Confirmed by the parallel Woodman Labs/GoPro chain where JPMorgan's release-by-secured-party was recorded 2021‑01‑25.

• 2025‑08‑04 (recorded) — Reel [not retrieved]

  • Conveyance: Security Interest / Security Agreement
  • Assignor: GoPro, Inc.
  • Assignee: Farallon Capital Management, L.L.C., as Agent
  • Correspondent: [not retrieved]
  • Context: Securitization (second lien) — collateral grant securing the $50M second‑lien term loan; Mateo Financing, LLC is the lender of record, Farallon the agent.

• 2025‑08‑04 (recorded) — Reel [not retrieved]

  • Conveyance: Security Interest / Security Agreement
  • Assignor: GoPro, Inc.
  • Assignee: Wells Fargo Bank, National Association, as Agent
  • Correspondent: [not retrieved]
  • Context: Securitization (first lien) — first‑priority security interest in all GoPro IP registrations/applications under the amended‑and‑restated Wells Fargo revolver.

Bottom line on the chain: there are recorded post‑issuance events, but every one is a security interest or its release. No assignment ever moved title away from GoPro. GoPro is still the owner of record.

---

Timeline diagram

timeline
    title Ownership of US 10574894
    2018 : Priority date
    2019 : Filed by GoPro Inc
         : Inventors assign rights to GoPro
    2020 : Patent issued
         : JPMorgan records security interest
    2021 : JPMorgan releases security interest
    2024 : GoPro asserts patent at ITC
    2025 : Farallon records second lien
         : Wells Fargo records first lien
         : PTAB upholds claims in IPR

---

NPE / troll-pattern signals

1. Shell-entity transfer — not present. No link moves title to any "IP / Holdings / Licensing / Ventures" entity. The only non‑GoPro assignees in the entire chain are a bank (JPMorgan), a lender/agent (Farallon), and a bank (Wells Fargo), and all three appear only as secured parties (2020‑10‑19, 2025‑08‑04 ×2). Title stays with GoPro.

2. Known asserter in the chain — not present. No assignee matches Acacia, Marathon, IV, IPNav, Wi‑LAN/Conversant, Vringo, Pendrell, Round Rock, Spangenberg entities, or any Unified/RPX high‑frequency plaintiff list. The 2024‑2025 litigation is GoPro suing a competitor (see below), not an NPE asserting purchased patents.

3. Repeat correspondent across the chain — unclear / not determinable. Correspondent names for this patent's recordings were not retrieved, so no recurrence can be confirmed or denied. (For context, GoPro's IP recordings have historically been filed by outside counsel at the Woodman Labs/GoPro Mountain View address; a specific repeat‑player attorney cannot be cited here without the reel/frame data.)

4. Cascading transfers — not present. There are no consecutive title assignments through chained LLCs. The 2020‑10‑19 → 2021‑01‑25 → 2025‑08‑04 sequence is a single owner (GoPro) layering and releasing liens, not a cascade of purchasers.

5. Pre-litigation transfer — not present. GoPro owned the patent continuously and asserted it in its own name: ITC Inv. No. 337‑TA‑1400 (complaint filed 2024‑03‑29, instituted 2024‑05‑06) and the parallel C.D. Cal. Case 8:24‑cv‑00681. The ITC's Final ID (2025‑07‑10) expressly states "GoPro is the owner by assignment of the asserted patents" — i.e., no assignment was staged before suit.

6. Bankruptcy fire-sale — not present. GoPro is an operating, NASDAQ‑listed company. The Aug 2025 secured financings are distress signals but not a Chapter 7/11 event and not a patent sale; IP was pledged as collateral, not conveyed.

7. Privateering — not present. GoPro is asserting directly against Insta360 (Arashi Vision Inc. / Arashi Vision (U.S.) LLC) — a market competitor — not funnelling the patent to an NPE to assert on its behalf. No secondary NPE plaintiff appears in the ITC or district‑court actions.

8. Defensive aggregator (anti‑NPE) — not present as a terminal assignee. The chain does not end at RPX/AST/LOT/Unified/OIN. Note for accuracy: IPR2024‑01186 was filed by Arashi Vision Inc. (d/b/a Insta360) as petitioner (instituted 2025‑01‑22; Final Written Decision 2025‑12‑10 finding no challenged claims unpatentable; now on appeal at the Federal Circuit, case 26‑1428) — that is a competitor's defensive IPR, not a Unified Patents challenge, and it does not affect ownership.

---

Verdict

Operating-company assertion.

Justification: The only recorded post‑issuance events (2020‑10‑19 JPMorgan security interest → 2021‑01‑25 release → 2025‑08‑04 Farallon second lien and Wells Fargo first lien) are security interests and a release, none of which transferred title; GoPro, Inc. remained the owner throughout, and the ITC's 2025‑07‑10 Final ID confirmed "GoPro is the owner by assignment of the asserted patents." GoPro ships products (HERO/HyperSmooth cameras) that embody the claims and asserted US 10574894 directly against a competitor, Insta360, in ITC Inv. No. 337‑TA‑1400 and C.D. Cal. 8:24‑cv‑00681. No shell transfer, no NPE in the chain, no pre‑litigation assignment — the only noteworthy item is GoPro's 2025 IP‑collateral financings, which are a corporate‑distress indicator, not an NPE signal.

Verification links:

• USPTO Assignment Center (search by patent number): https://assignment.uspto.gov/patent/index.html
• Google Patents legal events for US 10574894: https://patents.google.com/patent/[US10574894B2](/patent/US10574894B2)/en
• ITC 337‑TA‑1400 Final ID (ownership finding): https://www.usitc.gov/press_room/documents/notice_of_issuance_final_idrd_1400.pdf
• PTAB IPR2024‑01186 docket: https://www.docketalarm.com/cases/PTAB/IPR2024-01186/

Open data gap to close on a follow‑up pass: pull the reel/frame numbers and correspondent-of‑record for each of the five recordings above directly from Assignment Center (query "10574894"). If the JPMorgan/Farallon/Wells Fargo recordings share a single outside‑counsel correspondent, that is normal financing practice and would still not change the operating‑company verdict — but it is the one field this report could not verify.

I'll research US patent 10574894 and its cited prior art. Let me start with searches.

Let me get the specific citation lists from the patent record and the IPR proceedings.

Let me pull the exact "Referenced Cited" lists from the patent's Justia and Google Patents records.

Prior Art Analysis — US 10,574,894 B2 ("Systems and Methods for Stabilizing Videos")

0. Note on scope and confidence

I retrieved the patent's own text, its related-continuation front page, and — most importantly — the active post-grant proceeding against it, IPR2024-01186 (Arashi Vision Inc. d/b/a Insta360 v. GoPro, Inc.), whose expert declaration and papers identify the references actually asserted against this patent. That IPR source is the most reliable grounding for "most relevant prior art" for US 10,574,894 specifically.

Two honesty caveats up front:

1. The fetched Google Patents page for US 10,574,894 did not include a complete, verbatim "References Cited" front‑page list (the fetch was truncated before that section). The reference list below is reconstructed from (a) the IPR2024‑01186 record and (b) the front page of the closely related continuation US 11,025,824 B2, which shares this specification. I flag which is which.
2. Importantly, the live challenge against this patent is a § 103 obviousness challenge (Zhou + Kwatra + Shi), not a § 102 anticipation challenge. I address the § 102 framing you asked for below, but I must be explicit that, on the record I retrieved, no single reference appears to anticipate all elements of independent claim 1 — that is precisely why the petitioner combined references.

---

1. Patent under analysis

Independent claim 1 (element structure per IPR2024‑01186): an image capture device with a housing; an optical element guiding light within an optical field of view; an image sensor generating an image signal during a capture duration; a position sensor generating a position signal; an electronic storage; and a processor configured to (1[f]) determine an observed trajectory of the housing from position information, (1[g]) determine a capture trajectory "based on a look ahead of the observed trajectory" reflecting actual/virtual positions, smoother than the observed trajectory, such that stabilization creates a more stable view than using the entire image content, and (1[h]) determine orientations of the capture field of view (punch‑outs); (1[i]) generate video content from the punch‑outs. Dependent claims 2, 3, 6, 7, 12, 13, 17 add "user's intended motion" preservation (claims 2, 7, 12, 13, 17) and inter‑/intra‑frame motion handling (claim 6).

---

2. Most relevant prior art references

A. Zhou — primary reference (asserted in IPR2024‑01186)

• Full citation: U.S. Patent Application Publication US 2016/0006935 A1 (Zhou); granted counterpart U.S. Patent No. 10,911,677 B2 (Zhou), issued 2021‑02‑02. (Both appear in the cited‑references list of related GoPro patent US 11,696,029.)
• Publication date: 2016‑01‑07 (publication). Predates the '894 priority date (2018‑05‑18).
• Brief description: Video‑stabilization system using a look‑ahead buffer of video input frames. Per the IPR record, Zhou discloses "a look‑ahead technique enabled by buffering a number of video input frames before generating a first stabilized video output frame," uses a motion buffer (identified as element 210) holding motion data for past and "future" frames (‑N … +N), determines a smoothed trajectory versus the device trajectory, modulates stabilization strength based on estimated motion blur, uses gyroscope output (quaternions) against rotation, respects an overscan/visible‑frame constraint, and addresses motion blur/jitter including intra‑frame motion.
• Potential § 102 relevance: Zhou is the reference closest to a single‑reference disclosure and is the linchpin of the asserted ground. It is mapped to claim‑1 elements 1[g][i] (look ahead), 1[g][iv] (stabilization more stable than entire content), and to claim 6 (inter‑frame minimization / intra‑frame preservation) and elements of claims 1[a]–1[f]. However, the petitioner combined Zhou with Kwatra to reach the "minimize rotational velocity and rotational acceleration" limitations, which the patent owner argues Zhou does not teach — so Zhou alone does not appear to anticipate claim 1.
• Anticipation candidate: dependent claim 6, and elements of claim 1 (look‑ahead, smoothed trajectory, punch‑out/overscan). Not independently anticipating claim 1 on the retrieved record.

B. Kwatra et al. — secondary reference (asserted in IPR2024‑01186)

• Full citation: U.S. Patent No. 8,531,535 B1/B2 — Kwatra et al. (assigned to Google; the reference is styled "Kwatra" in the IPR). Also listed as "8531535 | September 10, 2013 | Kwatra et al." in the cited references of GoPro's US 11,696,029.
• Issue date: 2013‑09‑10 (well before the priority date).
• Brief description: Generating a smooth (stabilized) camera path by minimizing derivatives of the original camera path, via a cost function combining first derivative (velocity) and second derivative (acceleration); includes a proximity constraint "that requires the smooth camera path motion to preserve an original intent of the recorded video," and supports constraints on "range of change in zoom and rotation." Discloses that the original camera path includes rotational/angular movement.
• Potential § 102 / § 103 relevance: Kwatra is the reference mapped to claim 1's "path that minimizes a combination of rotational velocity and rotational acceleration of the housing" (IPR ¶¶107–112) and to claim 2 ("preserves a user's intended motion") via its proximity/original‑intent constraint. It is a § 103 combination reference, not an anticipatory reference — the IPR frames it as supplying the minimization teaching missing from Zhou.

C. Shi — third reference (asserted in IPR2024‑01186)

• Full citation: "Shi" (exhibit designated EX1006‑family in IPR2024‑01186; the exact patent/publication number was not retrieved in my searches).
• Brief description (per the case summary): Discloses intentional‑motion preservation in video stabilization.
• Potential relevance: Asserted in combination with Zhou + Kwatra to supply the "intended motion" limitations of claims 2, 3, 7, 12, 13, and 17.
• Confidence note: I could not verify Shi's full citation, number, or date from the retrieved results. Treat this entry as unverified; do not rely on it without confirming the exhibit number in the IPR2024‑01186 exhibit list.

D. Front‑page references on the related continuation US 11,025,824 B2 (same specification)

These were cited on the front page of the '894 patent's continuation and are part of the family's cited‑art record:

For US 6,654,019 and US 6,982,746, the retrieved snippets do not give enough claim‑level detail to assert which specific '894 claims they might anticipate; on the record they read as background/§ 102(b)‑era references rather than anticipatory art for claim 1. The WO publication dates above are approximate and should be confirmed against the actual documents, since I did not retrieve their bibliographic data directly.

---

3. Direct answer to the § 102 question, and the § 102/§ 103 caveat

• No retrieved reference appears to anticipate independent claim 1 under § 102. Zhou lacks the explicit "minimize a combination of rotational velocity and rotational acceleration" limitation; Kwatra supplies the minimization teaching but does not disclose the full image‑capture‑device‑with‑housing / look‑ahead architecture. This is confirmed by the fact that IPR2024‑01186 pleaded these references together under § 103, not singly under § 102.
• Best § 102 candidates (dependent claims):
  • Claim 6 (capture trajectory minimizing inter‑frame motion while preserving intra‑frame motion based on motion during frame exposure) → Zhou (look‑ahead buffer, motion‑blur/jitter and intra‑frame exposure discussion). This is the strongest single‑reference anticipation candidate in the record.
  • Claim 2 / claims 7, 12, 13, 17 ("preserves a user's intended motion") → Kwatra (proximity constraint preserving "original intent") and/or Shi; likely § 103 rather than § 102.
• § 103 obviousness (the live ground): Zhou (look‑ahead, smoothed trajectory, punch‑out/overscan) in view of Kwatra (minimization of motion derivatives / rotational velocity + acceleration, intent preservation) and further in view of Shi (intentional‑motion preservation) — targeting claims 1, 2, 3, 6, 7, 12, 13, and 17. The PTAB Final Written Decision in IPR2024‑01186 (noted in the patent's litigation data as issued) would be the authoritative statement of which claims, if any, were held unpatentable; I did not retrieve its outcome in these searches and will not speculate about it.

---

4. Items I could not verify (stated explicitly, not fabricated)

1. Complete verbatim "References Cited" list on the US 10,574,894 front page — not retrieved; reconstructed from the IPR and the related continuation US 11,025,824.
2. Shi's full citation (number, title, date) — not retrieved; the IPR case summary confirms its role but not its identity.
3. Exact publication dates for the three WO references — approximate, inferred from numbering conventions; confirm against the documents.
4. Outcome of IPR2024‑01186 (Final Written Decision) — the Google Patents litigation block shows an FWD was filed, but I did not retrieve its holding.
5. Whether any examiner‑cited art during original prosecution (as opposed to IPR‑asserted art) anticipates any claim — I did not retrieve the file‑wrapper "References Cited" list, so I cannot state the examiner's citations with confidence.

If you want, I can attempt a second pass specifically targeting the IPR2024‑01186 exhibit list (to pin down "Shi" and the EX1004/EX1005 identities) and the Justia front‑page "Referenced Cited" table for US 10,574,894 to replace the reconstructions above with primary citations.

I'll gather the prior-art record for the '894 patent, including the IPR grounds and cited references, then build the §103 analysis.

Let me pull the specific prior-art references (Kwatra, Zhou, Shi) and the citation record for the '894.

Let me identify the Zhou and Shi references precisely.

Obviousness Analysis — U.S. Patent No. 10,574,894 under 35 U.S.C. § 103

Prepared: April 26, 2026 · Patent: US 10,574,894 B2 ("the '894 patent"), Systems and methods for stabilizing videos · Assignee: GoPro, Inc. · Effective filing/priority date: May 18, 2018 (AIA case; pre‑AIA § 102(g)/§ 103 practice does not apply)

---

0. Sourcing, scope, and what "prior art" is actually before me

Two disclosure notes, because they materially affect the reliability of this analysis:

1. The supplied page text did not contain a rendered "References Cited / Cited By" listing. The Google Patents page as fetched gives the specification, the abstract, the litigation block, classifications, and the algorithmic "Prior art keywords" line (capture, trajectory, housing, images, view) — but not the examiner's citation list. So "the prior art on this page" that is actually usable is: (a) the examiner‑cited reference identified in the file history (Zhou), and (b) the prior‑art grounds of record in IPR2024‑01186 (Zhou, Kwatra, Shi), which is the only formally adjudicated §103 challenge to this patent.
2. The verbatim granted claim text is still not in front of me. Consistent with the caveat already flagged in the patent‑summary section, the limitation breakdown below is reconstructed from (i) the specification you supplied, and (ii) claim language quoted in the IPR papers (Petition, Patent Owner Response, expert declarations, and the Board's institution decision). Limitation labels such as "1[a]" are my own. Claim numbers for dependent claims are inferred and flagged as such.

Consistency check against earlier sections: the earlier "Patent summary" flagged (a) the IPR petitioner identity as unconfirmed and (b) the CAFC docket as unverified. Later searching in this session resolved both: the IPR petitioner is Arashi Vision (U.S.) LLC d/b/a Insta360, and CAFC No. 26‑1428 is Insta360's appeal (Notice of Appeal filed Feb. 6, 2026) of the Dec. 10, 2025 Final Written Decision in IPR2024‑01186. No other contradictions found. Nothing in this section depends on those points, however; the §103 analysis turns on the references, not the parties.

---

1. Legal framework applied

• Graham v. John Deere Co., 383 U.S. 1 (1966): scope/content of the prior art; differences between the prior art and the claims; PHOSITA level; secondary considerations.
• KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007): a combination is obvious where (i) elements are combined per known methods to yield predictable results, (ii) a known technique is used to improve a similar device in the same way, (iii) a finite number of identified, predictable solutions exists, or (iv) there is a design incentive/ market pressure.
• In re Kahn / KSR: the motivation must be articulated with a "rational underpinning," not a conclusory assertion.
• In re Fulton, 391 F.3d 1195 (Fed. Cir. 2004): mere preference is not teaching away; the reference must criticize, discredit, or discourage. This is the pivot point of this case (see §6).
• In re Magnum Oil Tools / Patent Owner's burden: once a prima facie case is made, the burden of production on teaching‑away and objective indicia shifts to the patent owner.
• Scope of analogous art: all references below are in digital video stabilization / image‑processing for handheld capture — the same field of endeavor as the '894 patent.

---

2. The prior‑art references

Zhou's material teachings (pin cites per the petition; Ex. 1004): a captured frame 305 (= the image with the optical field of view) containing a movable visible frame 300 (= the capture field of view / punch‑out) and an overscan 310 = 305 − 300 (the margin); a frame buffer holding (N+1) frames and a (2N+1)‑entry motion buffer so that stabilization of "current frame 0" uses both past and future motion data (col. 4:24–56) — i.e., an express look‑ahead; a smoothed motion sample computed from past/future data, with the correction being the difference between the motion data and the smoothed trajectory (col. 7:47–51); the corrected frame is kept "within its overscan" (Abstract; col. 7:24–38); motion data comes from gyroscopes/accelerometers and camera rotation/orientation is computed at integration start/end (col. 5:30–53); and the correction equations expressly contemplate "other axes" and gyroscope output, not translation alone (col. 7:24–38).

Kwatra's material teachings (Ex. 1005): estimate the original camera path C(t); compute a new optimal smooth path P(t) as an independent optimization (not a filter applied to C(t)); solve a crop‑window transform B(t); apply B(t) to recast the video. The objective is

O(P) = w₁|D(P)| + w₂|D²(P)| + w₃|D³(P)| — i.e., minimize a weighted combination of path velocity, acceleration and jerk (Eq. 7; spec. 13:23–14:12), with weights a, b, c that "steer the minimization" and an express 100:1 weighting between motion components (L1‑LP slack‑variable discussion), i.e. "weighing them equally or differently." Kwatra also discloses the very constraints the '894 recites generically: an "inclusion constraint" — the crop window must stay within the original frame rectangle (the margin constraint); a "proximity constraint" preserving the original intent of the movie; and a "saliency constraint" — salient points (e.g., from a face detector) must be included in the crop window, so that "our path centers the object of interest (jumping girl) in the middle of the crop window" (the target constraint).

Two points in Kwatra's favor that the petitioner under‑used (my observation, flagged for the reader):

• Kwatra's path is a generated smooth path, not a modified observed path. The '894 specification states the capture trajectory "may be determined by generating a smooth path that respects the set of constraints, rather than by modifying the observed trajectory." Kwatra's L1 optimization produces exactly that (constant / linear / parabolic segments), which blunts any argument that Kwatra is a mere low‑pass filter of the type the '894 disparages.
• Kwatra's saliency/face‑directed path maps almost verbatim onto the '894's "target constraint … determined based on positions of a target in the images … so that the target is within one or more of the punch‑outs."

---

3. Ground 1 — Zhou + Kwatra against independent claims 1 and 11

3.1 Limitation map (reconstructed)

Claim 11 is the apparatus counterpart and is met by the same combination plus Zhou's mobile‑device hardware (col. 12:12–21, Fig. 13).

Claim 5 (position sensor = gyroscope/accelerometer/IMU; position information independent of image information) is met by Zhou's inertial motion data.

3.2 Why a PHOSITA would have combined them (petitioner's articulation)

1. Same field, same problem, same mechanism. Both dampen camera shake in handheld video (Ex. 1004, 1:41–45; Ex. 1005, 3:58–59) and both operate by moving a crop window inside a larger captured frame (Zhou's visible frame 300; Kwatra's crop window). KSR rationale (i): known elements combined according to known methods → predictable result.
2. Complementary, not duplicative. Zhou modulates stabilization strength based on estimated blur but does not optimize the shape of the smoothed trajectory itself; Kwatra supplies exactly that missing optimization. KSR rationale (ii): known technique used to improve a similar device in the same way.
3. The constraint structures are mutually compatible. Kwatra's "inclusion constraint" (crop window must remain inside the frame rectangle) is the same physical constraint as Zhou's "keep the current frame within its overscan"; a PHOSITA would recognize that one is the mathematical expression of the other.
4. Finite number of predictable solutions. For path smoothing, minimizing the first and/or second derivative is a small, well‑known set (constant path / constant velocity / constant acceleration segments); Kwatra itself frames the approach as a generalization of L1 trend filtering, a known technique. KSR rationale (iii).
5. Reasonable expectation of success / implementability. Kwatra's Eq. 7 is an LP with slack variables — the petition's expert characterized it as "straightforward calculations" that a handheld device's processor can handle — and Zhou's device is a smartphone‑class platform with the gyro/accelerometer input Kwatra's transform consumes.
6. Design incentive. Consumer demand for smoother handheld footage; sensor‑based and hybrid stabilization pipelines were already known.

3.3 Where Ground 1 is weak (this is why the claims survived)

• The disputed limitation is narrow and load‑bearing. The only element not plainly in Zhou is the weighted minimization of rotational velocity and rotational acceleration of the housing. The petitioner's mapping of that element onto Kwatra requires reading Kwatra's path variable P(t) as rotational. GoPro's expert (Dr. Foroosh) argued Kwatra's P is a 2D image‑space linear motion model and that "one cannot simply replace P with theta."
• Express statement arguably amounting to teaching away. Kwatra's provisional (Ex. 2019, at 3, 5) states the all‑parameters joint estimation "requires a more complex non‑linear optimization method which is not fast enough for real‑time performance," and that "since imposing smoothness constraints directly on s and theta would be non‑linear, we instead constrain just a and b." A fact finder can read that as criticism/discrediting of rotational smoothing (→ In re Fulton teaching away), or merely as a stated design preference for a real‑time use case (→ not teaching away).
• Modality gap. GoPro argued that substituting a sensor‑derived trajectory into a cost function designed for image‑based linear models is not a mere substitution of known elements, and that the motivation as pleaded was "conclusory" (citing Teleflex v. Ficosa).
• Outcome of record. The Board's Final Written Decision of Dec. 10, 2025 determined no challenged claims unpatentable (PTAB docket: "Final Written Decision – Appealed"; appeal at CAFC No. 26‑1428). The parallel ITC investigation also found no violation as to the '894 and no infringement of the '894 by the accused products.

My assessment of Ground 1: as pleaded, it is a colorable but fragile prima facie case. The look‑ahead, punch‑out, overscan/margin, and sensor limitations are strongly met by Zhou; the equilibrium of the case rests entirely on (i) whether Kwatra's P is rotational and (ii) whether Kwatra's real‑time justification rises to teaching away. Both are fact questions on which the patent owner's evidence prevailed below.

---

4. Ground 2 — Zhou + Kwatra + Shi (dependent claims 2, 3, 7, 12, 13, 17)

Petitioner added Shi (a 3D sensor‑based motion‑estimation reference addressing preservation of intentional motion) to supply: (a) the characterization of the smoother trajectory as having less jitter, (b) the set of constraints framework, and (c) in particular the trajectory constraint — i.e., determining the constraint on the path from a subsequent portion of the observed trajectory so that intentional user motion is preserved.

Motivation, as pleaded: both Zhou and Shi are motion‑sensor‑domain references (3D/rotational), so Shi cures Kwatra's image‑domain limitation; a PHOSITA seeking to discriminate intended panning from unintended shake would look to the intentional‑motion literature.

Why Ground 2 largely collapses with Ground 1: the patent owner correctly observed that Shi is not alleged to cure the deficiencies of Zhou+Kwatra with respect to independent claims 1 and 11 (or claim 5). A reference cannot rescue a prima facie case on the independent claim merely by adding disclosure to dependent claims; if claims 1 and 11 are not shown obvious, the dependent claims fall with them.

Analyst note: Ground 2 is the right instinct — a sensor/rotation‑domain reference is exactly what's missing — but Shi is the wrong reference for the job (it goes to intentional‑motion classification, not to rotational path minimization). See §6 for references that would fill that gap far more directly.

---

5. Dependent claims — reconstructed mapping

⚠️ Claim numbers below are inferred from (i) the IPR's grouping of Ground 2 to claims 2, 3, 7, 12, 13, 17, (ii) the POR's identification of claim 5 as the position‑sensor claim, and (iii) the specification's enumerated implementations. Confirm against the granted claims before relying on any specific number.

If the "less jitter," "set of constraints," "margin," and "target" limitations are in fact distributed across the dependent claims as above, then Ground 1 alone already reaches the margin and target constraints via Zhou and Kwatra — GoPro's ITC termination of claims 2–4 and 6–20 and the Board's focus on claims 1 and 11 suggest the dependent claims were never the center of gravity.

---

6. Combinations that would make a materially stronger §103 case

These are my recommendations as an analyst; each reference must still be vetted for § 102(a)(1)/(a)(2) qualification (every one appears to predate May 18, 2018) and for analogous‑art status (all are).

Combination A — Zhou + Kwatra + Bell (my primary recommendation)

• Bell, Troccoli & Pulli, "A Non‑Linear Filter for Gyroscope‑Based Video Stabilization," ECCV 2014 (also relied on in the sibling IPRs IPR2024‑01164 / IPR2024‑01165 concerning the related '840 patent — so it is already familiar to this record).
• What it fixes: Bell supplies smoothing applied directly to gyroscope‑derived rotational data with a non‑linear filter — i.e., precisely the capability the patent owner says Kwatra lacks. It neutralizes the "Kwatra is limited to 2D image‑based linear models" argument, because rotation minimization arrives from a reference built for rotations.
• Motivation (KSR (ii)): same field (gyro‑based stabilization of handheld cameras), same input signal (gyro angular rate), same output goal (smoothed rotational trajectory used to crop frames). A PHOSITA would improve Zhou's sensor‑based smoothing with Bell's rot:ational filter in the same way Kwatra improves it with derivative minimization, and Bell's filter is applied to the same data Zhou already generates.
• Expectation of success: Bell reports working, implementable gyro‑based stabilization; coupling it to a crop‑window (Zhou/Kwatra) is a mechanical application of a known transform.

Combination B — Zhou + Kwatra + Karpenko (US 9,071,756)

• Karpenko discloses constraint‑based rotation smoothing run in a coarse‑to‑fine mode to improve convergence. This directly supplies (i) rotational smoothing, (ii) the "set of constraints" of dep. claims 3/13, and (iii) a motivation (computational efficiency) that is exactly what the patent owner says Kwatra's rotational smoothing cannot achieve. A petitioner who couples Karpenko's efficiency story with Kwatra's weighting story has an answer to the "not fast enough for real‑time" objection.

Combination C — Zhou + Cai (US 10,158,802) ± Kwatra

• Cai teaches distinguishing "zero motion" from "intended motion" by analyzing a sequence of future frames (e.g., the next five frames) — the trajectory constraint derived from a subsequent portion of the observed trajectory. Zhou already supplies the look‑ahead buffer; Cai supplies the semantics of intent. This is the combination the petitioner used against the sibling '840 patent (Zhou + Cai, Ground 1), and it maps more cleanly onto the '894's stated inventive purpose ("preserves a user's intended motion") than Shi does.

Combination D — Kwatra + Bell + Karpenko (Zhou or a sensor‑based primary optional)

• For any claim construction in which "combination of rotational velocity and rotational acceleration … weighing them equally or differently" requires an explicit weighted sum of first and second derivatives of a rotation signal: Kwatra supplies the weighted‑derivative objective and the crop‑window constraints; Bell and/or Karpenko supply the rotational application; Zhou (or Ovrch & Forssén, "Gyroscope‑based Video Stabilisation with Auto‑calibration," ICRA 2015) supplies the sensor‑derived observed trajectory and the look‑ahead.

Combination E — add a "margin/dynamic crop" reference for the margin constraint

• US 8,810,666 (Kwatra‑family, "using dynamic crop") iteratively selects the crop‑window size and a smoothness metric for the modified path — a direct teaching of computing the margin from the difference between full‑frame FOV and crop FOV, if the dependent claim requires the margin to be computed rather than merely used.

Why these combinations are more likely to succeed than Grounds 1–2

---

7. Secondary considerations (Graham factor 4) — and the nexus question

GoPro's objective‑indicia record, as recited in the Patent Owner Response, is substantial and, if credited with a nexus to the claimed feature, could independently defeat obviousness:

• Commercial success of HyperSmooth as implemented in the HERO7 Black (the POR asserts it "practices" the claim).
• Industry praise including an Emmy award for HyperSmooth stabilization.
• Long‑felt need and failure of others (the reference frame: low‑pass filtering and past‑motion prediction both failed to preserve user intent — the very problems the specification frames in connection with FIGS. 5A/5B).
• Copying by competitors.
• Unexpected results attributed to the simultaneous, differentially weighted minimization of rotational velocity and acceleration.

Nexus is the battleground. Objective evidence is only probative if tied to the claimed invention rather than to GoPro's commercial ecosystem. The claim feature that carries the case is the weighted rotational‑velocity/acceleration minimization; GoPro's evidence is aimed exactly there, which is why the Board's non‑obviousness determination is defensible given the record as pleaded. A stronger primary reference set (Combination A or B) would force the patent owner to re‑establish nexus for that same feature against a more on‑point teaching.

---

8. Bottom line

• Ground 1 (Zhou + Kwatra) and Ground 2 (Zhou + Kwatra + Shi) are the only §103 combinations formally of record. As pleaded they establish obviousness of everything except the "minimizes a combination of rotational velocity and rotational acceleration … weighing them equally or differently" element, which is carried entirely by Kwatra and is vulnerable to the arguments that (i) Kwatra's path is 2D/image‑based and (ii) Kwatra's specification arguably teaches away from non‑linear rotational optimization. On this record the Board determined no challenged claims unpatentable (Dec. 10, 2025), now on appeal (CAFC 26‑1428).
• The claims are not invulnerable. The look‑ahead, sensor‑independence, punch‑out/overscan (margin), and target/saliency limitations are squarely disclosed by Zhou + Kwatra (with saliency/face‑detection in Kwatra mapping onto the "target constraint"). The gap is narrow and reference‑specific.
• The strongest untested routes are those that source the rotational smoothing from a sensor/rotation‑domain reference rather than from Kwatra: Zhou + Kwatra + Bell (ECCV 2014), Zhou + Kwatra + Karpenko (US 9,071,756), Zhou + Cai (US 10,158,802), and Kwatra + Bell + Karpenko with a sensor primary. These pair the weighted‑derivative objective and crop‑window constraints (Kwatra) with explicit rotational filtering (Bell/Karpenko), removing both the modality gap and the teaching‑away foothold.

Confidence: high on the reference identities, the Zhou claim‑chart mappings, and the procedural posture; moderate on the precise phrasing/ordering of the claim limitations and low on the claim numbers of the dependent limitations, because I could not retrieve the granted claims verbatim.

---

9. Verification gaps (do not rely on these without checking)

1. Verbatim claim text of claims 1–20 was never retrieved; the limitation breakdown is reconstructed. Pull the granted claims from USPTO PatentCenter.
2. Identity of Shi. I could not confirm which Shи reference was Pet. Ex. 1006–1008 in IPR2024‑01186. That reference's identity and content must be confirmed before Ground 2 can be assessed on the merits.
3. Column/line pin cites for Zhou and Kwatra are quoted from the petition/POR excerpts surfaced in search, not from my own reading of the documents. Verify against the copies of US 9,674,438 and US 8,531,535.
4. Kwatra provisional (Ex. 2019) language ("not fast enough for real‑time performance") is quoted from the Patent Owner Response. This is the fulcrum of the teaching‑away question and must be read in context.
5. Objective‑indicia record (Emmy, HERO7 Black sales, copying evidence) is recited from the POR only; it has not been independently verified, and no nexus finding has been made by a fact finder that I can confirm.
6. Not auto‑corrected: US 9,674,438 ("Zhou"), US 8,531,535 and US 8,810,666 (Kwatra family), US 9,071,756 (Karpenko), US 10,158,802 (Cai), US 9,464,676/US 9,294,676 (Apple "choosing optimal correction"), and the '894 patent itself are all distinct documents and have been treated as such.