IVS Briefs
IVS / Edge AI Research Brief
Overall assessment
Signal level
High on last night’s Monroe unanimous 23 and Monterey 3–2 / 10-year kill, Cottage Grove’s ALPR-ban ordinance, and the ROSCon pair (Isaac ROS 5.0 + Qualcomm/MoveIt).
Notable trends
- Kill-path is now unanimous suburb (Monroe), sheriff+DA lose a 10-year (Monterey), and ban-the-category (Cottage Grove) in the same 48h.
- Texas freeze: Plano walks away; Dallas still backfills.
- Flock’s keep-slide this week is a Bishop Fox PDF, not a new camera.
- Edge news is Isaac ROS on Jetson + MoveIt under Qualcomm, not another locked dome.
Gaps
No new Genetec / Milestone / Axis VMS SKU (a&s Adria Genetec SaaS and Pelco interoperability are reprints). No new Hailo/DeepX/Jetson module SKU — Isaac ROS 5.0 is software on existing Jetson. Lebanon 09-22 follow-up has no vote primary. College Station / New Bedford / Dallas 321 are this week’s clocks, not yet votes.
Mix counts
LPR 3 / VMS 0 / Edge 2 / Policy 1 / Lab 1 (plus Other 1 Bishop Fox). LPR+Policy city/contract = 4/8 = 50%.
Now vs On deck
Now 6 (items 1–6) · On deck 2 (PickNik, LEAP-NBV).
Signals (8)
- 1
Key point
Monroe City Council voted unanimously to end the Flock contract and deactivate all 23 ALPRs. Mayor and vice mayor were absent (neither voted). City started Flock in 2024 on a two-year, $149,950 deal. Residents split: veteran Debra Ridenhour (“digital dragnet we never consented to”) vs William Michael (no reasonable expectation of privacy on plates in public). Councilmember Michael Graves: not an indictment of police; “fear of big government.” Molly Cloyd: rollout skipped real community choice. Todd Hickman surprised it was unanimous. Path to bring cams back exists but not in time for November.
Why it matters
Second Greater Cincinnati suburb after Lebanon’s Sept 8 pause. Quote unanimous + 23 + $150k on every OH RFP. Lebanon’s Sept 22 follow-up still has no vote primary this morning.
Source
WKRC Local 12; WCPO - 2
Key point
Board voted 3–2 not to let the Sheriff’s Office negotiate a new 10-year Flock deal covering 63 existing cams at ~$1.7 million (walked back from 103 / ~$2.8M). Supervisors Luis Alejo and Chris Lopez voted to keep talking. Sheriff Tina Nieto and DA Jeannine Pacioni urged yes. Chief ADA Berkley Brannon: pass a state misuse crime, “don’t throw the baby out with the bathwater.” Prior contract expired July 25; Flock had been running month-to-month. Board also: write ALPR guidelines and explore county-owned readers (staff said expensive, unlikely).
Why it matters
A sheriff + DA keep-slide lost. The close was no 10-year lock, not a new SKU. If CA RFPs still want ALPR, bid county-owned + statute-misuse, not another decade of Flock.
Source
KSBW; Monterey Herald - 3
Key point
Village Board voted unanimously to immediately stop Flock, deactivate or cover the readers, terminate the contract, and remove hardware. Trustee Paula Severson brought it. Unlike Sun Prairie / Fitchburg (shopping Axon or another vendor), Cottage Grove directed staff and the village attorney to draft an ordinance banning the village and PD from purchasing, leasing, contracting, installing, operating, or reactivating ALPR. Dane County roundup the same day: only four departments still on Flock; Cottage Grove had 9.
Why it matters
This is a technology ban, not a vendor swap. Do not bid a replacement LPR in Cottage Grove until that ordinance dies. Wisconsin keep-path is now Axon-or-nothing in some towns and no ALPR at all in others.
Source
WMTV; WKOW; Channel 3000 - 4
Key point
Plano turned off 90+ Flock cams after Gov. Greg Abbott ordered state agencies to stop funding ALPRs. Chief Ed Drain: “game-changer,” used ~three years, funded by a ~$300k/yr Texas Anti-Gang grant that ended with the order. Unlike Dallas (321 grant cams on a Sept 25 10-day clock, still hunting city money), Plano is not seeking alternative funds for now. Council also hearing privacy pushback on drone-as-first-responder.
Why it matters
Same freeze, two buyer postures: Dallas backfill, Plano walk away. Quote Anti-Gang grant dead → 90+ dark, no replacement RFP when a Texas city asks what peers did.
- 5
Key point
Independent Bishop Fox “open/clear box” test of hardware (ALPR, PTZ, gunshot, Compute Box), software (Nova, Safety Platform, FlockOS, DFR), mobile apps, AWS, and all source. 36 issues: 2 Critical and 7 High fully remediated and independently verified; 16 Medium (13 done, 2 partial — not remotely exploitable without inside access); 1 Medium not remediated (Android 8.1 on first-gen Falcon — Flock says it backports itself, no hardware refresh); 3 Low done. 24/36 fixed. Notable fix: Android Verified Boot / Secure Boot on older LPR so stolen poles can’t boot untrusted images and dump the local stills. Flock: no customer data accessed by anyone outside Flock in this test. Full report for customers at security.flocksafety.com. Fixes deploy centrally — no agency patch.
Why it matters
Every Flock keep-RFP will now ask for this PDF. Put Secure Boot on legacy poles + Falcon 8.1 residual in writing. Do not treat “pen-test done” as “hardware is current.”
Source
Flock / Chris Castaldo - 6
Key point
Isaac ROS 5.0 ships GPU-accelerated ROS packages with agent skills (open Agent Skills format: env setup, NITROS→rosidl::Buffer migration, FoundationStereo fine-tune, pick-and-place). Moves to ROS 2 Lyrical + Ubuntu 24.04; NVIDIA contributed a CUDA buffer backend so nodes can zero-copy GPU payloads. FoundationPose agent-ready inference: object pose up to 5.5× faster (NVIDIA claim). isaac_ros_cumotion partner bench (Ekumen): collision-free arm paths 2–5 ms. Runs from Jetson Orin Nano to Jetson Thor. Users named: Intrinsic, Mentee, EKUMEN, Flexiv, Universal Robots AI Accelerator, Seeed reBot on Thor. Open source, no license fee. NITROS-direct callers need a source-level migration.
Why it matters
This is the Jetson perception stack, not a locked AI dome. Spec Isaac ROS 5.0 + Orin/Thor, not a camera brand’s 36-month model. Accelerated compute beats locked “AI cameras”: swap models vs a 36–84 month hardware refresh.
- 7
Key point
Qualcomm Technologies agreed to acquire PickNik Inc. (Boulder; steward of open-source MoveIt manipulation). Terms undisclosed; customary closing conditions. Plan: keep MoveIt 1 and 2 open-source under the existing license, keep third-party hardware support, and streamline MoveIt on Dragonwing robotics SoCs (and Arduino) so developers go from VLA/VLM → planning → real-time control. Nakul Duggal (Qualcomm) and Dave Coleman (PickNik founder/CPO) on the “open robotics / physical AI” line. Sits next to Qualcomm’s Arduino (Oct 2025) and Dragonwing IQ10 (Jan 2026) stack.
Why it matters
AMR / gate-arm RFPs will ask if motion planning is locked to one SoC. Quote MoveIt stays OSS + Dragonwing is optional. Accelerated compute beats locked “AI cameras”: swap models vs a 36–84 month hardware refresh.
- 8
Key point
Can a foundation-model 3D Human Mesh Recovery loop plan the next-best view on-board when you cannot offload? LEAP-NBV distills large HMR teachers into a 32M student (offline mesh objective), FP16-quantizes the vision encoder, and runs the closed loop on NVIDIA Jetson Xavier NX. Distillation recovers 6–7 mm PA-MPVPE vs the undistilled student. Edge-optimal config: HMR engine ~12 ms; full loop 3.6 FPS, 2.6 J/frame — 2.0× faster and 3.0× lower energy than uncompressed, with near-matching downstream quality.
Why it matters
Tactical / contested-link sites will ask for on-box NBV, not a cloud VLM. Keep the Jetson, shrink the model. Accelerated compute beats locked “AI cameras”: swap models vs a 36–84 month hardware refresh.