Landroid Vision, neural networks and personal information
Vision and data privacy
Today data privacy is more important than ever, which is why we want to make sure our customers know exactly how Positec handles the data acquired through Landroid Vision’s camera.
First of all, Landroid Vision’s camera inherently respects privacy considerations. Angled downward at 38 degrees, its field of view is restricted to the ground approximately 140 cm in front of the mower. As a result, the likelihood of the camera capturing personally-identifiable data is exceptionally low.
Landroid Vision works without a boundary wire thanks to the implementation of a neural network. What’s that? Neural networks are computer systems designed to mimic the way the human brain learns. They learn from examples and experiences, which allows them to identify patterns and make decisions. For instance, a neural network can learn to identify grass after seeing millions of grass pictures. This process is called machine learning.
Landroid Vision’s neural network is trained through images to distinguish grass from anything else - fences, pathways and other garden features. This means that in the rare case that Vision collects images containing items that are not grass related, such as toys, pets or license plates, it does not categorize them, and it just classifies them as “non-grass”. If Vision acquires images with this kind of content, the picture undergoes a blurring and anonymizing process to ensure that all personally-identifiable data is removed.
Under normal operating circumstances, Landroid Vision only captures images if its confidence level is low. What’s a confidence level? When Vision finds itself in an environment it doesn’t easily recognise, its confidence level can decrease thus potentially complicating the decision-making process (to cut or not to cut? that is the question). Maybe Vision is being confronted with a type of ground it hasn’t been trained for, or maybe it’s just not easily recognisable. In such cases, Vision takes pictures so that it can learn from its experience and improve its own performance. This process is essential to all machine learning systems.
That said, every image we collect goes through an anonymization process. Let us go a little bit more into the details.
Images and data are collected by Vision for two purposes:
- Product improvement: this data is completely anonymized and it’s only used for machine learning purposes. No data linked to the customer is kept. Data such as time, place and robot mower identifiers are removed and images can’t be linked to their user in any way.
- Diagnostics: image content is still anonymized, however data related to time and identification can be used by our customer service to provide users with the best assistance possible.
When images are collected from your Landroid Vision they are uploaded to secure cloud servers located in the same geographic region as the mower(for example, for a mower in the EU, the images are uploaded to a secure cloud server in the EU). Here data undergoes an automated anonymization process - it blurs out possible personally-identifiable content such as faces, license plates, and house numbers, and deletes images-related data (such as location and time of the shot). The resulting images no longer contain personal data and are therefore completely anonymized. The anonymized images are then transferred to the machine learning server.
A few more words on machine learning. Images must be labeled before being fed to the neural network. What does “labeling” mean? In the case of Vision’s neural network, someone must tag an image as “grass” or “not grass”. This process is managed by a third party on Positec’s behalf, a common practice in machine learning training. It’s important to note, though, that the images are already anonymized before they are handled by the external contractor. By the time the images are received by the external contractor, they do not contain any personal information or sensitive data anymore.
In compliance with GDPR, customers can choose to opt out from data sharing via the App at any time. To opt out, simply open the Menu on your Landroid app, go to Settings, Logging and disable diagnostic information for troubleshooting and/or product improvement.
Positec takes your privacy very seriously. We do not and never will sell or rent your personal data to third parties. Please see the Landroid Privacy Notice Supplement (which sets out details about how personal data is collected and used within the context of the app).
If you have any questions, please do not hesitate to get in touch on the contact details listed below.
For issues related to data privacy, please contact us at datenschutzeu@pisitecgroup.com.