Bosch has developed a self-learning AI sensor for personalising wearable and hearable devices.

Self-learning and personalisation features help users easily train devices with customised fitness activities. The aim is to reduce development time, cost, complexity and power consumption with automatic tracking.

The edge AI requires no cloud connectivity, reducing latency and improving privacy.

Squats, crunches or kettlebell swings, the list of fitness exercises available nowadays is almost endless. Many fitness trackers or smartwatches, however, offer only a very limited number of activities that can be tracked and do not recognise activities reliably due to different performance styles, equipment, heights and energy levels of users.

To solve such problems, Bosch Sensortec has developed a self-learning motion sensor that adds artificial intelligence (AI) to portable devices.

The BHI260AP self-learning AI sensor enables manufacturers of wearable and hearable devices to provide personalised fitness tracking through self-learning AI software in the sensor. It recognises and adapts to a wide variety of movements and can learn any new fitness activity that is based on repetitive, cyclical patterns. Users can hence be trainers and trainees at the same time.

“The self-learning AI sensor will change how users interact with their fitness devices from a mere one-way approach to an interactive way of training,” said Stefan Finkbeiner, CEO at Bosch Sensortec. “This new sensor combines Bosch Sensortec’s long-term experience in smart motion sensors with its strong competence in innovative software development.”

The self-learning AI software is available with a standard set of more than fifteen pre-learned fitness activities, so no training is required before use. In addition, it offers four product features: learn, personalise, auto track and enhance.

The learning mode offers users the option to add fitness activities that were originally not supported, enabling them to customise the device to their needs. The personalisation feature lets users adapt existing, pre-learnt activities to their own individual style, increasing the accuracy of calorie counting and activity analyses.

With the auto track function, users can automatically track fitness activities without any manual intervention and analyse their intensity with activity type and count over time, enabling both endurance and strength training.

Finally, manufacturers can add fitness activities without having to modify the software or needing an original dataset. These transferable exercises may be provided by coaches or star athletes, enabling benchmarking against the best and learning from experts, or simply from the users’ friends. This enhances the perceived value of devices and helps manufacturers differentiate.

With the AI running on the sensor itself, hence on the edge, no connectivity to the cloud or even a smartphone is needed. This keeps the data private and means activities can be continuously tracked and analysed, without the need for an internet connection or tethering to a phone. Edge AI also reduces latency and power consumption, meaning users can get fast, real-time feedback and extended charging intervals on their devices.

The new sensor is provided to manufacturers as an integrated system in package (SiP), which includes the hardware, software and embedded AI. This reduces development time and costs, and cuts time to market.

In addition to the self-learning AI software, Bosch offers other software for the BHI260AP. Wearables manufacturers can simply load the required software on to the sensor to enable applications targeted at specific use cases including orientation tracking, position tracking and swimming. The sensor recognises four different swimming styles – butterfly, backstroke, breaststroke and freestyle.

Since the BHI260AP is a fully programmable sensor, manufacturers can build customised software to be embedded in or upload customised software depending on the users’ context. To protect this customised software from unauthorised use, the smart programmable sensor has a digital signature.

To make development of AI based wearable solutions easier, Bosch offers an application board along with a BHI260AP shuttle board, which can be wirelessly connected over Bluetooth Low Energy to smartphones.

The self-learning AI sensor and the corresponding shuttle board are available via Bosch distributors. The different software packages are available for download at the Bosch Sensortec web site.