A New Fitness Tracker Sensor Can Learn and Recognize Almost Any Activity

Depending on the sport, different athletes rely on different exercises during their fitness routines and while fitness trackers have gotten smarter, they can generally only recognize a handful of motions like swimming, riding a bike, or running. Bosch has developed a new self-contained chip that can be taught to recognize any activity using AI, so an athlete could train their fitness tracker to detect and record every individual exercise in their routine.

Fitness trackers started out as nothing more than step counters, but eventually gained technologies like GPS and sophisticated gyroscopes to let joggers and runners know how far and fast they went. Eventually, as their processors became more capable, they gained the ability to recognize the difference between swim strokes versus the motions of riding a bike, allowing more accurate metrics to be collected during a workout.

For athletes looking to improve their performance, studying those metrics can be critical, but cycling vs. swimming is still a vague distinction. There are countless different strokes a swimmer can make, for example, and most fitness trackers don’t take the different forms into account while they record laps across a pool. If you’ve got a couch on the sidelines it’s not a problem, but most swimmers don’t, which has led Bosch Sensortec to develop a new motion sensor for fitness trackers that uses artificial intelligence to learn very specific motions and automatically recognize them later.

The BHI260AP self-learning AI sensor might not sound very motivational (remember the cooler sounding Nike Fuelband?) but it promises to make fitness trackers an even more useful tool during training. It’s small enough to incorporate into wearables and even wireless headphones, and out of the box Bosch includes a set of more than 15 different pre-learned fitness activities. That’s a good start, meaning many athletes can just jump right into a workout, but adding to that database of recognizable activities is as easy as putting a device into a learning routine and going through the repetitive motions a handful of times. Those pre-learned activities can also be tweaked for a user’s given technique or style, which ensures that the data collected, and other metrics calculated, like calories burned, are more accurate.

The motion sensor features artificial intelligence built right in, so the data doesn’t have to be bounced through a connected smartphone and up to the cloud to be analyzed. The AI-powered learning happens all in the chip which means the data should be private and an athlete shouldn’t have to keep their phone within reach during a workout.

Once a new activity has been learned, users also don’t have to manually select it from a list before starting. The motion sensor is smart enough to recognize an activity it already knows, and automatically start stat-tracking for a very specific routine. Instead of getting a report that a user lifted weights for 20 minutes, they’d get a detailed breakdown of every curl or bench press they performed. Because the sensor can be trained on a per person basis, it could potentially even alert a user when their form is slightly off compared to the ideal they specified when training the AI. Could fitness trackers replace coaches all together?

There are currently no consumer-ready devices featuring Bosch’s new BHI260AP self-learning AI sensor built-in, but the hardware is now available to developers and OEMs so assuming 2021 isn’t as chaotic as 2020 was, there’s a chance we may be seeing this functionality added to popular fitness trackers in the coming year.