1 Characteristics and compatibility of BHI260AP
1.1 General information
- What is the overall size of the package?
The size of the system-in-package with an accelerometer, gyroscope and a programmable 32bit microcontroller is 3.6 mm * 4.1 mm.
- Where are you in the commercialization phase - i.e. Prototype? Sampling? Beta customers?
The mass production launch is in preparation.
1.2 MCU
- Is there a MCU inside the BHI260AP?
Yes, there is a MCU inside.
- Can the raw data from the BHI260AP be accessed in real-time and does the sensor include a multi-axis accelerometer?
Yes, the raw sensor data can be accessed in real-time. The sensor includes a 3-Axis accelerometer and a 3-Axis Gyroscope.
- Is the BHI260AP compatible with other MCUs?
Yes, the BHI260AP can be used together with other MCUs
1.3 IMU
- Is there any downside or potential tradeoff to having a 6-axis versus a 9-axis IMU, or is drift not an issue in this application?
The recommendation of 6-axis or 9-axis is dependent on the typical usage environment, whether it is subject to magnetic distortion in indoor and/or outdoor environments. While the current magnetometers suffer from hard-iron and soft-iron effects, the sensor is equipped with a state-of-the-art sensor data fusion that includes a fast magnetometer calibrator to help reduce or control drift in direction after magnetic distortions are detected. Here, an absolute generic recommendation is not possible due to the nature of sensors, but guidance can be provided specific to the application and expected use. Insights on this topic are available in our publication (EDN whitepaper)
1.4 Power management
- What is the typical mean power consumption when the sensor is steadily in use (e.g. when someone is doing a workout)?
It depends on the number of sensors activated inside the sensor (accelerometer or gyroscope) and the variety of activities simultaneously being tracked. For an exact number, please get in touch with Bosch Sensortec.
- What is a possible strategy to manage power efficiently, and is ultimate power consumption determined by the customer developing the wearable?
Typically, for saving system power, the developers look for power-expensive components such as the GPS, display, connectivity and the running time of application processors, etc.
1.5 Sensor fusion and ML:
- How are you using sensor fusion to develop information from the raw sensor output? Please provide an example.
- Sensor data are fused together for multiple purposes. One of them is the generation of relative and absolute orientation. This orientation information is quite useful for tracking the angle (heading, pitch and roll) at which the device moves in relation to the user's body movements.
- Individual sensors, i.e. an accelerometer, gyroscope, magnetometer and pressure sensor suffer from problems such as distortion and sensitivity caused by different environmental and operating conditions, such as temperature. Sensor data fusion helps to leverage the strengths of one sensor to correct the output of the other sensors, and vice-versa.
- With a reliable estimate of heading over longer periods of time, the position tracking and navigation becomes reliable even when GPS is running in duty cycle mode to save system power.
- Similarly, sensor data is fused together to classify and count user functions such as swimming and fitness tracking, as individual sensors (accelerometer or gyroscope) do not generate unique enough information to identify the type of movement.
-
- Can you explain the different ways that sensor fusion, ML, and a powerful software framework are being used in this sensor and the difference between them?
- This smart sensor system offers many state-of-the-art features that are very useful for devices worn on the wrist or head. These include dedicated solutions for pedestrian dead reckoning, swimming analytics and fitness tracking, as well as generic software platforms with self-learning AI software and orientation estimation.
- The sensor also comes with a software development kit that allows you to program and add your own code to these platforms and dedicated solutions. This makes creating solutions with the sensor easier.
- The orientation estimation platform can be used to develop applications such as position tracking, while the self-learning AI platform can be used to develop applications that involve tracking repetitive patterns during use for a simplified human-machine interface or to enable a personalized recommendation system.
- The software framework makes it possible to offer all functions as virtual sensors and thus activate/deactivate them depending on the context of the end user.
2 AI firmware and algorithm of the BHI260AP
- Is the AI firmware updateable?
Yes, the AI firmware is updatable.
- What is the prediction accuracy rate for the trained algorithms?
The typical prediction accuracy is >90% for personalized pre-trained activities.
- Which neural network topology is used in AI? Or is it just a statistical analysis for pattern recognition?
It is a proprietary patented Bosch technology designed especially for learning on the edge (which enables personalization) and still keeps computational and power requirements low.
- What is the classification based on, is there a NN used?
Classification (learning and recognition) is based on time-series accel and gyro data using a proprietary patented Bosch technology designed especially for learning on the edge (which enables personalization) and still keeps computational and power requirements low.
- What types of software algorithms have you developed and what are their functionalities?
Self-learning AI software for fitness tracking
This software enables on-device learning and automatic tracking of a wide variety of fitness movements, including options for on-device individual-specific personalization of movements and support for increasing numbers of activities without the need to modify the original software.
Swim analytics
This software dedicated for wrist wearables generates useful information of users’ swimming activities, such as length count, style of swimming and stroke counts.
Pedestrian dead reckoning
This software helps to reduce the power consumption of a wearable device by enabling the duty cycling of power-consuming GNSS components, as well as improving the accuracy of outdoor positioning with pedestrian dead reckoning.
Relative and absolute orientation
This software estimates relative and absolute orientation of the device, including outputs such as rotation vector, game rotation vector, linear acceleration and gravity.
3 Applications and use cases of the BHI260AP
- Some of the heavy compound exercises might be dangerous if they are not done correctly. Is there a measure to prevent incorrect training due to the incompetence of the users?
For providing guidance to the user for appropriate training, you can make use of a combination of the recognition and orientation tracking included inside BHI260AP.
- Is the sensor able to indicate that the workout is done correctly?
To provide guidance to the user for appropriate training, you can use a combination of recognition and orientation tracking included in the BHI260AP.
- The ability to indicate if a workout is done correctly would open up the field to physiotherapy applications, right?
Yes, correct.
- How is the software reacting to highly variating movements like “burpees directly followed by heavy deadlifts” etc.? Is it picking up those constantly changing movements properly performed in a row when learned individually?
Yes, the sensor is picking up the changing movements automatically, without any explicit information from user.
- How configurable is the BHI260AP for different activities? For example, a swimming expert with biomechanics background may have some different focus than the built in swimming analysis. Will developers be able to configure that?
Yes, developers would be able to create and load custom patterns for the activities of their interest. Please contact us via the Bosch Sensortec Community if you need further support on this topic.
- Can the self-learning AI Software be used for other activities than fitness, like for e.g. fishing?
Yes, please get in touch with us via the Bosch Sensortec Community for further discussions on how to achieve this.
- This product has a lot of functionalities - what is the most innovative aspect/feature/function?
The most innovative aspect is the self-learning function designed inside the sensor itself, which enables a seamless personalization of activity tracking for a wide variety of activities and end-users.
4 Tools for BHI260AP
- Do you have development boards, like for e.g. a pulse band?
Yes, we have application and shuttle boards. Shuttle boards for BHI260AP are not yet available from our distributors. We plan to make them available in Q2 2021.
- Do you provide a design kit and a free graphical user interface (GUI) with real-time raw data display at Windows OS?
Yes, we provide software development kit for programming the BHI260AP and the PC based development desktop tool for visualization of real-time data.
- Does the BHI260AP need any real-time cloud AI support or is it an edge AI sensor?
BHI260AP is a complete edge sensor. Cloud support is not needed.
- Is there an IDE available to use the API and customize it?
Yes, we provide a software development kit for programming the BHI260AP and a sensor API for customizing the sensor.
- Is it going to be a black box for the devs or will it be an open software development kit?
We will provide a software development kit for programming the BHI260AP and make it as open as possible for developers.
- Is there any data analytics previewed in real time (smart data profiling) or that are used for diagnostic predictions for instance?
Yes, data analytics can be previewed in real time with the development desktop.
