The Pelvar harness
Give it a technology area. A claim-family scope. A named human reviewer who will sign the output. Ninety minutes from input to draft.
Cognex Corporation — Machine Vision Competitive Landscape
Draft v0.1 · agent-generated · 2026-06-27 11:42 EDT · Pre-review · not yet certified
Executive summary
Cognex Corporation (NASDAQ: CGNX) holds a granted-patent portfolio of approximately 480 active families across machine vision, deep-learning-based defect detection, and industrial barcode reading. Three claim families dominate strategic relevance for the company's next-cycle R&D bet: (a) 3D-vision-with-deep-learning defect detection, (b) edge-deployed inspection inference, and (c) cross-modal vision-LIDAR fusion. The competitive frontier is bifurcating between deep-learning-first entrants and traditional machine-vision incumbents with bolted-on AI modules.
Three claim families, ranked by strategic relevance
3D-vision deep-learning defect detection
Strongest concentration of recent granted activity (last 36 months). Cognex's portfolio shows broad claim coverage on convolutional-network architectures applied to point-cloud defect classification. Key family root: US 11,XXX,XXX. Continuation strategy active — three CIPs filed Q4 2025.
Granted families: 47 · Pending: 12 · Forward citations (3y): 211
Edge-deployed inspection inference
Growing claim emphasis on quantization, model-distillation, and runtime constraints for on-camera inference. Cognex is competing with embedded-vision specialists (Basler, Allied Vision) and inference-accelerator startups in this space.
Granted families: 31 · Pending: 18 · Forward citations (3y): 96
Cross-modal vision-LIDAR fusion
Earlier-stage claim concentration. Smaller granted base, but rapid pending-application growth signals an active scouting position rather than a defensive one.
Granted families: 9 · Pending: 14 · Forward citations (3y): 22
Competitor categorization
Agents categorized the top 8 cited entities into direct / adjacent / scouting positions. Two entries (highlighted) are likely mis-categorized and will be flagged in reviewer markup.
| Entity | Category (raw) | Concentration |
|---|---|---|
| Keyence Corporation | Direct | High |
| Omron Corporation | Direct | High |
| Sick AG · | Adjacent | High |
| Basler AG | Adjacent | Medium |
| Allied Vision Technologies | Adjacent | Medium |
| Landing AI · | Scouting | Low |
| Zivid AS | Scouting | Low |
| Photoneo s.r.o. | Scouting | Low |
Two competitor entries flagged for reviewer attention (amber dots). The raw draft is a competent first pass; the certified delta between this draft and the signed final is what tier-2 OEMs pay for.
Reviewer markup — Karen Presley, JD, LLM, MBA, CLP
Review session · 2026-06-27 14:08 EDT · Three corrections · two re-categorizations · Eval-set signal captured
3D-vision deep-learning defect detection
Cognex's portfolio shows broad claim coverage on convolutional-network architectures applied to point-cloud defect classification.
Reviewer: Cognex's portfolio shows broad claim coverage on specific point-cloud preprocessing pipelines feeding convolutional defect classifiers. The claim scope is narrower than "any CNN architecture for point-cloud defect" — it is gated on a defined preprocessing step (US 11,XXX,XXX claim 1, element b). This distinction matters for FTO analysis on adjacent entrants.
Sick AG — adjacent → direct
Sick AG · Adjacent · High concentration
Reviewer: Sick AG · Direct · High concentration. Sick's Inspector and PIM60 lines now ship with embedded deep-learning inspection. The "adjacent" categorization was based on legacy product categorization; current product strategy puts them in direct competition on the same buyer accounts.
Landing AI — scouting → adjacent
Landing AI · Scouting · Low concentration
Reviewer: Landing AI · Adjacent · Medium concentration. Landing AI is not pursuing the same patent strategy as Cognex (they are software-first, lighter portfolio), but their commercial overlap with Cognex's deep-learning inspection product line is real. Categorization should reflect commercial threat, not patent-portfolio shape.
Edge-deployed inference family
Reviewer: Two of the top-5 prior-art cites in the edge-inference section are over-weighted by the agent. They are foundational references rather than relevance-critical. Re-ranking applied. The agent's relevance scoring tends to over-weight forward-citation count for foundational patents; tracked as an eval-set signal for future calibration.
Three corrections captured. Each is labeled signal for the proprietary eval set. After 100 shipped briefs the eval set becomes a structural advantage that horizontal tools cannot replicate because they do not own the delivery.
Cognex Corporation
Machine Vision & Deep-Learning Inspection Portfolio
We compress two weeks of consulting work into ten days of signed, defensible intelligence. This brief was drafted by the Pelvar harness on June 27 at 11:42 EDT, marked up by the named reviewer above between 14:08 and 14:24 EDT, and certified for release at 14:32 EDT — three hours and twenty-six minutes of human reviewer time over an end-to-end engagement window of five business days.
Executive summary
Cognex Corporation holds a granted-patent portfolio of approximately 480 active families across machine vision, deep-learning-based defect detection, and industrial barcode reading. Three claim families dominate strategic relevance for the company's next-cycle R&D bet: (a) 3D-vision-with-deep-learning defect detection, gated on a defined preprocessing pipeline rather than any CNN architecture; (b) edge-deployed inspection inference; and (c) cross-modal vision-LIDAR fusion. The competitive frontier is now direct-competitive against Sick AG (re-categorized from adjacent), with Landing AI as a commercially-adjacent software-first entrant.
Strategic recommendations
The Cognex board should treat the 3D-vision deep-learning preprocessing family as the strategic core of any next-cycle R&D bet; continuation strategy is active and three CIPs filed in Q4 2025 should be supported through prosecution. Edge-deployed inference is a parity space rather than a moat space; resource allocation should match commercial demand rather than chase claim coverage. Vision-LIDAR fusion is at the right scouting stage for a 24-month build-or-acquire decision; one or two boutique acquisitions in this segment would compress the timeline by 18 months.
What changed in the certified brief vs. the raw draft
- Claim scope on 3D-vision family narrowed to the preprocessing-pipeline element of the lead claim.
- Sick AG moved from adjacent to direct competitor.
- Landing AI moved from scouting to adjacent.
- Prior-art re-ranking on edge-inference cites; two foundational refs down-weighted.
This is page one of a 27-page certified deliverable. Subsequent pages contain claim-family deep dives, prior-art tables, competitor profiles, and the in-licensing target shortlist. Full brief available to the Cognex VP of R&D at the engagement reference above.