Deebot Robot Vacuums Are Using Photos and Audio to Train Their AI (2025)

Deebot Robot Vacuums Are Using Photos and Audio to Train Their AI

An Australian news agency is reporting that robot vacuum cleaners from the Chinese company Deebot are surreptitiously taking photos and recording audio, and sending that data back to the vendor to train their AIs.

Ecovacs’s privacy policy—available elsewhere in the app—allows for blanket collection of user data for research purposes, including:

• The 2D or 3D map of the user’s house generated by the device
• Voice recordings from the device’s microphone
• Photos or videos recorded by the device’s camera

It also states that voice recordings, videos and photos that are deleted via the app may continue to be held and used by Ecovacs.

No word on whether the recorded audio is being used to train the vacuum in some way, or whether it is being used to train a LLM.

Slashdot thread.

Tags: artificial intelligence, China, cybersecurity, data privacy, Internet of Things, privacy, robotics

Posted on October 10, 2024 at 7:00 AM•6 Comments

Paul •October 10, 2024 9:22 AM

I wonder if the data collection can be switched off and still have the vac function.
Perhaps it’s very hackable like most consumer electronics, maybe burglars and law enforcement will use it to map a house?
Maybe the data collected will be sold to advertisers?

Sean •October 10, 2024 10:56 AM

Most likely being used to gain information about the users, likely to “better plan cleaning for quiet times”, but also likely also being used by “others” to gain things like passwords, business conversations with sensitive potentially useful information, especially as they are marketed at people who are likely living with disposable income, and who are WFH at large companies, where this info can be useful to competitors, and of course the layout of the house, including the address, makes it a great way to gather info about potential business espionage.

Whayt could go wrong, when even the furniture, along with the smart TV, is gathering data all the time, totally unfettered, about whatever is there, with decent audio quality and video good enough to be able to tell twins apart, and also with the potential for blackmail always being present as an almost free addition.

lurker •October 10, 2024 1:00 PM

Do these things work when there is no internet? Why does a vacuum cleaner need internet? Sadly, there will be people who think the last thread about GGlass was scary, but are happy to have these things trundling round their floor.

Those of us old-fashioned enough to do our own cleaning with our own hands will observe this is just The Future saying “hi”. As @Clive noted in the previous thread, Neil Stephenson saw it coming thirty years ago.

Winter •October 11, 2024 4:40 AM

No word on whether the recorded audio is being used to train the vacuum in some way, or whether it is being used to train a LLM.

There is some information about the claimed use in the article and they say they collect it only from users who opted-in to the/a program:

In a blog post from 2020, two engineers in the Ecovacs Robotics AI department described a problem they’d been facing.

“Building a deep learning model without large amounts of data is like making a house without blueprints,” wrote Liang Bao and Chengqi Lv.

“Due to the unique ground-view perspective and uncommon object categories, we cannot find any public dataset which fit our needs.

“Therefore, we first cooperated with many institutions to collect data from all over the world.”

But since the products have been launched, they confirmed that data from users who had opted into its “Product Improvement Program” was being used for training its AI model.

“During this data collection, we anonymise user information at the machine level, ensuring that only the anonymised data is uploaded to our servers,” the spokesperson said in a statement.

“We have implemented strict access management protocols for viewing and utilising this anonymised user data.”

The blog post mentioned is here:
‘https://blog.tensorflow.org/2020/01/ecovacs-robotics-ai-robotic-vacuum.html

Although it does not say anything about the data send home now, the idea is to use the trained model offline to protect privacy:

The robotic vacuum cleaner works at home where privacy is a priority. In order to protect users’ privacy and prevent data leakage, we need to deploy the neural network model on the robot in offline mode and use hardware accelerator to speed up inference as fast as we can.

Anon 456 •October 11, 2024 7:56 PM

I have yet to buy a robotic vacuum and/or mop, but I do admit they look pretty convenient for automated upkeep in between manual deep cleanings. That’s cool. Same with robotic lawnmowers that are coming down in price. The privacy implications of IoT do worry me, though.

Years ago I’d read some creepy things about how iRobot (maker of the popular Roomba robot vacuums) created detailed maps and floor plans of our homes but then uploaded it to company servers for G*d knows what, and that turned me off. Coupled with Amazon’s acquisition of iRobot, I quickly found myself looking to alternate brands like Shark and others.

Ecovacs seem to get pretty good reviews for their feature set. It’s unfortunate that it’s yet another IoT company based in a country known to be hostile to user privacy. I had no idea until today.

Back to the drawing board. Might lean back towards Shark if I decide to dip my toes in robotic vacuums and moppers.

Specific to this article, the audio recording feature is especially creepy to me. No way in heck would I allow any IoT or “smart” home device to possess a microphone.

James Clapper testified in 2016:

“”Smart” devices incorporated into the electric grid, vehicles—including autonomous vehicles—and household appliances are improving efficiency, energy conservation, and convenience. However, security industry analysts have demonstrated that many of these new systems can threaten data privacy, data integrity, or continuity of services. In the future, intelligence services might use the IoT for identification, surveillance, monitoring, location tracking, and targeting for recruitment, or to gain access to networks or user credentials.”
( https://www.armed-services.senate.gov/imo/media/doc/Clapper_02-09-16.pdf )

I’m torn between balancing the convenience of some smart home devices with my strong desire for privacy and peace of mind for my family. Our smartphones are already our biggest voluntary “bugs;” I don’t need to consciously add more.

iAPX •October 12, 2024 1:50 PM

@Winter

“During this data collection, we anonymise user information at the machine level, ensuring that only the anonymised data is uploaded to our servers,” the spokesperson said in a statement.

This made me really suspicious (as the typos) !
You can only anonymize through a third part, to hide originating IP Address at least and mix ENCRYPTED data from different sources, only knowing the source IP Address without access to the data.

There’s nothing anonymous in a WiFi connected device sending datas, including PII into a remote server, and nothing private as the servers are under Chinese control.