Creating interactive and engaging products means recognizing people and familiar objects. A toy or household robot designed to greet and meet the needs of its owner must first be able to learn and then recognize the owner. It also needs to be able to learn and store familiar objects and places.
However, privacy is a major concern. Many people do not want to share information across the Internet where it might be stored on a central server and mined for information. Specific privacy laws may also make learning and recognition via the Internet illegal. For example, in the United States, the Children’s
Online Privacy Protection Act (COPPA) gives parents control over what information an online service can collect from their kids.
Neurala’s Brains for Bots software is so efficient that learning and recognition can take place locally on the device. The Neurala brain can work even with low cost hardware, such as the Orange Pi Zero, that are appropriate for many toys.
Real-time, on-device learning can be set to learn a person or any arbitrary object. On devices without a screen, The Neurala Brain can be taught to learn the first person it sees. On devices with a screen, the end user can simply select an object by drawing a bounding box around the object. All devices can be shipped with several pre-taught objects, such as common household objects.
Real-time, on-device recognition can be set to alert the product when the desired object is seen. For example, a toy can be set to respond when it recognizes its owner. With a more powerful processor, a product can also find the object and inform the toy of its on-screen location.
Products using Neurala wouldn’t need to be connected to the internet or send information to remote servers, so private information or images […] stay on the device and nowhere else."