top of page

Over the edge with wearables

Why AI should no longer rely solely on cloud or mobile devices

The basic building blocks of wearables (sensors, processors and wireless connectivity) are evolving rapidly, partly driven by our evolving understanding of the data processing requirements of wearable technology products, but more fundamentally by silicon providers recognising the opportunity and establishing dedicated product lines in the space.

Wearables tend to be very small in size, low in weight and are almost always connected to the wider internet through the airwaves, using technologies such as Bluetooth or WiFi. They exhibit very limited storage capacity and the energy cost of transmitting data can be extremely high.

The trend a few years ago with IoT device data architecture was to ‘ship it all to the cloud’ where massive compute capacity could gobble up the data and present findings back to the edge nodes if required. However, this approach is crude, and although there is generally a need for a certain amount of data transmission, a stream of low-bandwidth insights, or ‘inferences’ is a much better model.

The edge of the edge

What we end up with is a class of IoT devices operating on the ‘edge of the edge’. These are devices that cannot be described under the normal edge computing terminology, which is usually used to describe mobile devices, industrial IoT devices and distributed racks of computers with semi-local resources. Wearables operate ‘over the edge’ and are extremely resource constrained. They require consideration on these terms, where data transmission relates directly to ‘time between charges’.

AI across a diversified set of technologies is driving power efficiency

What is of primary interest to us, as designers of wearable products are the new technologies coming online for local, ultra-low-power AI processing. The benefits of processing local data on the edge device are amplified when considering that the mobile phone is a general purpose device. Although Apple’s A12 and other new mobile AI innovations are tailored to low-power machine learning, they are predominated by voice and video signal processing applications and, as noted above, they are only available at the other side of an expensive Bluetooth link. This can also introduce serious issues with processing latency and the potential for the wearable to drop out if the phone is unavailable. Not great in a biofeedback system operating in real time. 

Figure 1 – An example of the AI resources available across the three major processing platforms of a general wearable technology ecosystem.

Traditionally the industry has used general-purpose processing in wearables, allowing signals to be gathered, filtered and passed on to the mobile device. However, there is now a new generation of processors available, or arriving on the market soon, with exceptional low-power processing capabilities. These AI-centric devices offer in the order of 10 or 100 times power efficiency and coupled with a saving of another 10 times through reducing wireless transmission, the opportunity for significantly improved user experience with wearables.

In addition to the new processing hardware, we are seeing continuous improvement and big steps forward in sensor designs, where the raw data can be pre-processed extremely efficiently, right at source by dedicated pre-processing. We will see the sensor designers, (such as Bosch, ST and others) continue to increase the intelligence and versatility of their devices, allowing us to develop extremely power-efficient products.

The future is intelligent wearable technology

In the early part of the decade, wearables were not meeting expectations in the marketplace. User experience and value was low, resulting in high abandonment and many examples of products failing in the marketplace. Since then, we have seen extraordinary advances in the underlying technologies and wearable technology opportunity is being magnified by this performance jump, facilitated by edge processing, insight generation and significantly improved user experience.

We are now at a tipping point, where there is an opportunity for AI-powered wearable technology products to finally start providing compelling solutions across the whole wearables space. Smart, connected products can now operate with extraordinary intelligence, creating deep health insights and user value, as well as immersive technologies, and responsive, subtle user experiences.

bottom of page