Free home cleaning in exchange for robot training data raises new questions
According to Ars Technica, a startup working in AI and robotics is sending camera-equipped cleaners into homes to collect robot training data in exchange for free cleaning. The approach is a household-environment extension of the 'human-camera data collection' model that has spread in recent years.
Under the programme's terms, a homeowner who signs up admits a worker trained by the startup at scheduled times. The worker films during the cleaning process using a head-mounted camera while walking through the home. In return, the homeowner pays nothing for the cleaning.
According to the startup's founders, the aim is to teach robots basic household tasks by observing human behaviour inside homes. As Ars Technica reports, most existing industrial robot training sets are based on factory or warehouse environments; that is not enough to train for the variety of furniture, carpets, household items and daily routines found in homes.
Data privacy specialists are cautious about the approach. Ars Technica relays a comment from Misha Rykov, lead analyst at the Mozilla Foundation's Privacy Not Included programme; Rykov said that 'recording activities that take place inside the home onto someone else's server is a data exposure that is difficult to reverse'.
The firm's privacy policy states that during filming the homeowner's face, sensitive documents and personal screens are automatically removed from the recording. The company says there is no sharing of data with third parties and that data is used only in robot training models. Ars Technica notes that a third-party audit report has not been published.
The startup's training sets cover human-performed tasks across different rooms of the home: kitchen cleaning, bathroom maintenance, washing dishes, folding clothes, sweeping carpets. The data forms the basis of training algorithms that engineers then apply to home robot prototypes in subsequent steps.
On the sector side, the home robot market has been growing rapidly in recent years. Three major companies – Tesla, Apptronik and Figure – are testing home robot prototypes in different segments. As Ars Technica notes, the quality of training data stands out as a determining factor in how these prototypes mature for market.
For the public, one of the critical questions is the purposes for which data collected inside the home may later be used. Where which items are stored, which cleaning products are preferred, what time the daily routine takes place in the home: these are data fields that open evaluation areas beyond robot training.
Eva Galperin, a representative of the Electronic Frontier Foundation (EFF), told Ars Technica that 'the purposes for recording in-home behavioural tracking data, retention periods and sharing limits should be presented in a way the homeowner truly understands'. That adds a new dimension to informed-consent debates in the sector.
This article should not be read as direct advice for legal, privacy or consumer decisions. The piece is limited to summarising the programme, sector data and privacy expert views reported by Ars Technica; anyone considering joining the programme is encouraged to review contract terms and data policies with a qualified legal expert or privacy advisor.