
In this issue
Can a robot remember where you left your wallet, and who moved it?
🤖 The pitch: Amazon's new patent describes a home robot that turns every object it sees into plain-language notes, so you can ask where your wallet is and get a sentence back.
⚠️ What's new?: What is new is the two-tier memory and a change log that can record which person was associated with the change.
💰 Follow the money: Amazon bought Fauna Robotics seven weeks after its first product launched, on US$16.6 to US$30 million raised.
📄 The paperwork: Ring's face recognition is already blocked in three states and facing a class action. Does a log naming who moved your wallet count as biometric data?
AI made PMs faster. Multiplayer mode is still broken.
A PM can summarize research, draft a PRD, and mock up a prototype before lunch. The hard part starts when the team has to decide what actually gets built.
Jira Product Discovery gives product teams one place to capture insights, prioritize ideas with consistent frameworks, and build living roadmaps stakeholders can rally around.
And because it’s connected to Jira, the context behind every decision stays with the work—so developers and their agents know not just what to build, but why.
AI helps PMs move faster. Jira Product Discovery helps the whole team build with confidence.
The 8:03 Search Party
You are already late. Keys, phone, coat, and then your heart totally sinks when you can't find your wallet. You check the kitchen counter, then the bowl by the door, then the counter again.
The robot has been standing in the hallway this whole time. So you ask it.
"The wallet is in the office, far corner of the desk, behind the stack of books."
If it knew where the wallet was, it knew when the wallet got there. It knew the thing sat on the counter until 8:03 last night. And if somebody carried it into the office, there is a decent chance it knows who.
A patent published in August 2026 describes that system in full. The applicant is a two-year-old New York startup, Fauna Robotics and the assignee is Amazon.
| HOW IT WORKS |
Fauna Robotics builds Sprout, a 3.5-foot, 50-pound humanoid with a soft shell and wiper-blade eyebrows, sold at around $50,000 to research labs. CNBC reported Amazon's acquisition of the company in March 2026, seven weeks after Sprout launched. CTO Josh Merel came out of Google DeepMind, and CEO Rob Cochran ran product at neural interface startup CTRL-labs before Meta bought it, per TNW.
The patent covers how a robot like that builds a memory of a house. It starts with a map of the house. As the robot walks, a camera and an inertial sensor feed a fast loop that pulls out keypoints, the distinctive corners and edges software can recognise again later, while a slower loop stitches small chunks of 3D space into one map of the home.
Each time the robot re-scans a room, it compares what it sees against what the map says was there. A difference means an object arrived, moved, or left.
When it spots one, it derives size, colour and shape, then converts them into plain words with a confidence score attached, like "wallet," "dark brown," "personal belongings."
Those words go into two databases. A short-term one holds what is in the house now. A long-term one holds the last known state of everything the robot has ever seen, alongside an event log of what changed and when.
The words are the whole trick. Storing "brown cardboard box near the entryway" costs a sliver of what the footage costs, which is how a small robot carries months of household history onboard. Ask where you left your wallet and the question runs through short-term memory, then long-term, then the event log, and comes back as a sentence.
When a stored object moves, the robot writes a change-log entry with a timestamp. The patent says that entry can also specify whether an identified person was associated with the change.
So yes, you will be able to blame your lost snacks on your siblings :)
| THE PROBLEM |
Home robots navigate well and don't remember a whole lot. A robot vacuum knows the shape of your living room. It cannot tell you the keys were on the arm of the couch on Tuesday.
That gap is why household robots stall at single-step chores. Ask one to fetch the laundry basket and it locates the basket from scratch, every time. Errands with three steps need a picture of where things live that survives between sessions.
The consumer cost is small and relentless. A Pixie survey, which is a tracker company's marketing as much as research, put it at a whopping 2.5 days a year per American hunting for misplaced things.
| WHO'S SOLVING IT? |
The category is household spatial memory, of giving a machine a persistent, askable record of the objects in a home.
1X Technologies is the closest match. Its NEO humanoid ships to US homes at $20,000 or $499 a month, with more than 10,000 units reserved (The Robot Report). NEO accumulates household knowledge too, but its central bet is human teleoperators logging in to finish tasks the robot cannot, with memory as the by-product. Fauna's approach takes the human out and hands the job to a database.
Robot vacuum makers already map homes and identify objects, and the better ones now have arms, but their memory is a cleaning map with obstacles on it rather than a searchable history. Any could add recall.
Ring keeps a labelled event history of your front door, including its Familiar Faces recognition feature, and Astro has been invite-only at $1,599 since 2021.
Defensibility in this area is a little shaky. Much of the perception stack is public research, and the filing cites it openly, including DISK for keypoints, MixVPR for place recognition, LightGlue for matching, FAISS for search. The novel claim is narrower, the two-tier object memory and the routing of a question through it.
| THE MARKET |
MarketsandMarkets puts the 2026 humanoid forecast at US$5.41 billion and 2035 at US$50.27 billion. Goldman Sachs models US$38 billion by 2035, while Morgan Stanley runs a US$5 trillion scenario for 2050 in which roughly 90% of the work is industrial.
Amazon has over 600 million Alexa-enabled devices in the world, the benchmark TechCrunch used to show how slowly Alexa+ was reaching people. Those devices sit in rooms, they have microphones, and more of them have cameras every year. A household object memory does not necessarily need to be deployed in a $50,000 humanoid to be commercially interesting. It does need a good visual scope of your home.
| DEAL FLOW |
Robotics capital is chasing bodies and brains. The home is being picked up on the side at a discount, and Amazon is doing most of the picking.
Start with what this patent cost. Fauna raised US$16.6 million according to Tracxn, while co-founder Josh Merel's own posts cite US$30 million from Kleiner Perkins, Quiet Capital and Lux Capital. Either way, Bloomberg broke the acquisition of a two-year-old company with 50 staff and one product.
Silicon Republic noted Amazon also bought Rivr, a Zurich startup making a stair-climbing delivery robot, the week before, and that both deals landed after Amazon cut staff in its own robotics division and halted its Blue Jay warehouse arm.
Crunchbase News reported that Apptronik, which builds the Apollo industrial humanoid, closed a US$520 million Series A extension in February 2026, pushing one round past US$935 million with Google, Mercedes-Benz and John Deere involved. Skild AI, a robot foundation model company, raised US$1.4 billion at a US$14 billion valuation. Figure AI sits at a reported US$39 billion. Humanoid startups raised roughly US$6.1 billion across 139 deals in 2025 on PitchBook data compiled by Cervo Technology, against 13,000 to 18,000 units actually sold.
Mordor Intelligence puts the smart tracker market at US$0.81 billion in 2026, growing to US$1.27 billion by 2031, with Apple and Samsung taking close to half the revenue. Tile, the company that invented the category, sold to family-locator app Life360 for US$205 million in 2021, according to TechCrunch.
Ring launched Search Party in September 2025, an AI feature that scans neighbours' outdoor cameras for lost dogs described in the app. By early 2026 it was reuniting more than one dog a day, had opened to people who own no Ring camera at all, and came with US$1 million to put Ring systems into 4,000 animal shelters.
| THE RISK |
Picture the ordinary versions. A teenager gets asked why the wallet moved at 8:03. A cleaner is queried about a missing ring on the strength of a probabilistic match. A house guest is logged for months and never told.
Amazon announced Ring's Familiar Faces in September 2025 and shipped it that December, after the EFF and Senator Ed Markey asked the company not to. Ring disabled it in Illinois, Texas and Portland, Oregon, and a nationwide class action followed in June 2026 from a Virginia man who says his face was scanned while visiting friends (Recording Law).
For the imagery that is kept, there is already a precedent. MIT Technology Review documented how development Roombas sent household photos to labelling contractors, and how images from inside homes ended up in private social media groups. Amazon says it brings decades of experience earning customer trust in the home. That experience also includes a US$30.8 million FTC settlement in 2023 over Alexa and Ring privacy failures…
Would you let a home robot keep a log of who moved what? |
| WHAT'S NEXT? |
Perhaps soon there will be a humanoid Alexa answering "where are my keys?" because something in your house has been keeping a list, and the list has timestamps and names on it. And it can even blame it on the person who displaced it!
That could point to a memory layer sitting under Amazon's existing devices, with a robot arriving later to act on what it already knows. It also raises the possibility of a household record that outlives the household, queryable by whoever holds the account and by anyone who can get access to it.
This week's patent is US 2026/0228231 A1, titled "Method for Spatial Mapping and Event Detection in Dynamic Environments", published by Amazon Technologies, Inc.
Hit reply and tell us whether you would put one in your hallway, or come argue with us on Instagram and LinkedIn.
| FOR THE NERDS |
• LightGlue: Local Feature Matching at Light Speed with ICCV: Explore one of the four public models the filing leans on, and see how much of this perception stack is open research rather than proprietary invention.
• Why iRobot went bankrupt with WBUR: Listen to co-founder Colin Angle explain what killed the first company to put robots in millions of homes, before betting on the second.
• The Roomba that recorded a woman on the toilet with MIT Technology Review: Follow a single household image from a test device to gig workers on three continents, and see why "we only store descriptions" deserves interrogating.
• The Humanoid Economy with Morgan Stanley: Zoom out to the US$5 trillion 2050 scenario, then note how little of it the bank assigns to households.
• Robotics venture funding in 2026 with Crunchbase News: Track where the capital is actually landing this year, and how far the consumer home sits from the centre of it.


