I’ve been a little frustrated with the state of the connected security market for a while now. After a spate of products such as Canary and Piper, which reconfigured the alarm system market with point products that use sensors integrated into a single hub to determine if a break-in has occurred, the market somewhat stagnated.
We still have a cluster of sensors on our doors, windows, and walls aimed at detecting when something opens, closes, moves, or breaks something inside the home. Amazon’s launch of the Alexa Guard glass break and smoke alarm detection feature as part of the Echo device represented a leap forward to that smarter security system, but it felt like a small leap.
Enter Minut, a five-year-old company based in Malmo, Sweden. The firm has built a device it calls Point that uses sensors and machine learning to determine if someone has broken into your home or to simply to let you know when something is wrong. Its most interesting feature is that it has done this in a way that protects the users’ privacy, even as it sends data to the cloud.
Nils Mattisson, co-founder and CEO of Minut, explains that the company began as a way for Airbnb hosts to keep an eye on their properties without having to keep a camera on their guests. The ability to tell if a guest has arrived, if something breaks, or if people are moving around in a home after guests were supposed to check out gives hosts peace of mind.
The first generation of the product used Wi-Fi and was mainly geared for notifications to a smartphone. A second generation added an alarm feature and faster notification times as customers started buying it as a security system for their own homes and rental properties. And soon a third generation aimed at mobile operators will add a cellular Cat-M and NB-IoT connection that will provide cellular backup.
So about those privacy features. Mattisson explains that privacy was important from the beginning, which is why the device never had a camera. It relies on sensors (motion, temperature, and sound) and machine learning to figure out what is happening in the home. But it considers privacy around the sensor data and machine learning as well. Instead of sending recordings of actual sounds to the cloud, Minut does machine learning on the device to figure out what types of data elements matter. It can also act on those elements at the device level to sound an alert.
However, if the elements are deemed novel, the device can send those elements to the cloud back-end to train the machine learning model further. Then new versions of the algorithm containing the new elements are sent down to the device. Mattisson does a good job explaining how it works: “You can’t go back to the raw sounds, and so we can preserve privacy and still use the full power of the [cloud] back-end,” he says. “We’re much better off doing things this way because the device still can recognize behavior offline and it ensures privacy. I think that will become more valuable in this space.”
He’s right. And even if privacy isn’t an issue, local processing of spoken commands or other automations are a trend that even Google is buying into. At Google I/O this year the company showed off a local software development kit and machine learning functions for the smart home related to voice.
As a side note, the frequent updates thanks to better-trained models also mean that Point can consume more data than I originally anticipated. Mattisson says the updates can require a few megabytes of data per day, which seems like a lot for mere sensor readings. Mattisson says this is why Cat-M, with its greater bandwidth, is a better radio for Point as opposed to NB-IoT.
This led me to consider the various ways companies should be thinking about connected smart devices. I’ve encouraged companies to add more memory to their products so they can handle security updates and new features, but it’s possible that companies should also think about how to future-proof radios to handle bandwidth needs associated with machine learning models and frequent updates.