Learn About Inertial Sensors with a Machine Learning Core

Edge computing is a method of combining and processing data on sensor platforms to generate information locally. New challenges, however, crop up with the integration of sophisticated heterogeneous sensors spewing out diverse data in substantial quantities. Prior calculations must be fed into these sensors to generate usable data or purposeful information. Furthermore, integrating the sensors into real applications requires coordinated development from both software and hardware teams, who need a common framework for condensed development times. These sensors are mostly connected to a cloud or fog where the Machine Learning (ML) algorithm or actual processing is performed on data sluiced in from the sensors nodes.

Machine Learning Core (MLC)

ML is an application of Artificial Intelligence (AI) where a machine learns by itself or under supervision, without explicit programming. The system acquires the ability to learn and progress from experience sans compromising data accuracy spontaneously. The ML processing capability permits the transference of several algorithms from the host processor to Inertial Measurement Units (IMUs). Technological advancements enabled ML cores to be embedded on a chip sprinkled with multiple sensors (e.g. motion sensors).  ST extends its motion-detecting machine-learning core (MLC) technologies application towards industrial and consumer applications through its ISM330DHCX, LSM6DSOX and LSM6DSRX 6-axis iNEMO Inertial Measurement Units.

The integrated sensors (gyroscope and accelerometer) sieve real-time motion data before it is sent to the Computation Block, where statistical parameters designated as "features" are applied to captured data. The features deposited in the computation block find use as inputs for the third block of the MLC.

The Decision Tree examines statistical parameters and contrasts them against a few particular thresholds to identify specific situations and the generated results piped to the MCU. The MLC results are the output results of the decision tree inclusive of the optional meta-classifier.

Figure 1: Machine Learning Core Blocks

The MLC performs necessary AI motion data pre-processing, using approximately from 1 to 1000 times lower power than an average MCU would consume to finish an identical task. Consequently, IMUs with this IP attribute can jettison the host MCU, allowing longer battery runtime, reduced maintenance, and diminished size and weight in motion-sensing and context-aware devices. Decision trees are created quicker and rapidly modernized using ML as compared to explicit programming when pertinent data sets can be accessed.

Motion Sensor Applications

Gyroscopes stop objects from falling over. A few mono-trains use gyroscopes. Boats or ships frequently use them so that the vessel remains the right way up in choppy seas. Gyroscopes also find use in monitoring patient movements inside healthcare systems.

ST has designed a 3 iNEMO inertial module that includes the ML core LSM6DSOX, LSM6DSRX, and ISM330DHCX. This is a system-in-package with a characteristic 3D digital accelerometer and a 3D digital gyroscope. The LSM6DSOX enhances performance at 0.55 mA in high-performance mode, and empowers always-on low-power features to amplify the consumer's optimal motion experience. The consumer-grade LSM6DSRX holds a 3-axis accelerometer and a 3-axis digital gyroscope with an extended full-scale angular-rate range up to ±4000dps and leading-edge performance in temperature and time. The LSM6DSRX and industrial-grade ISM330DHCX with embedded temperature compensation for superior stability arrives with a 10-year product-longevity assurance and is -40°C to 105°C specified, .

All three inertial modules support main OS requirements and offer real, virtual, and batch sensors with 9 kilobytes for dynamic data batching. ST's family of MEMS sensor modules leverages the robust and mature manufacturing processes already in use for micro machined accelerometers and gyroscope production. Specialized micromachining processes manufacture various sensing elements, while CMOS technology develops IC interfaces allowing a customized circuit design trimmed to suit the sensing element attributes.

Figure 2: Machine Learning Core in the LSM6DSOX, LSM6DSRX, and ISM330DHCX.

This module encloses an MLC and a programmable finite-state machine logic (FSM) that allow motion data classification based on known patterns. Such unique features mitigate the principal processor from the initial activity tracking stage, conserving energy to accelerate motion-centric applications. Devices provisioned with the LSM6DSOX deliver a responsive and agreeable "always-on" user experience without sacrificing battery runtime.

The device's internal memory exceeds conventional sensors.  The ultra-modern high-speed I3C digital interface enables brief connection times and prolonged periods between main controller interactions, resulting in higher energy savings.

The MLC in each device links to the integrated FSM logic and runs simple repetitive algorithms, such as hits, counting steps, or rotations at a lower power compared to a microcontroller. The FSM signals the principal controller after it detects a preset occurrence of events or after a specific elapsed time.

Development Boards and Software,

The new inertial sensors are best tested by acquiring their respective development boards. The LSM6DSRX, ISM330DHCX, and LSM6DSOX and correspond to STEVAL-MKI195V1 and STEVAL-MKI207V1, respectively.  Both match with the STEVAL-MKI109V3STEVAL-MKI109V3 motherboard, commencing on the STM32F401VE microcontroller, and are the fastest way to experiment with the Unico GUI and initiate work on a prototype.

Table 1: IMU Sensors with corresponding Evaluation kits

UNICO-GUI is a complete software package for all MEMS sensors' evaluation boards available in the ST product portfolio (like accelerometers, environmental sensors, gyroscopes, and magnetometers). It is accessible for Mac OS X, Windows, and Linux Debian-based platforms.

UNICO-GUI enables a quick sensor setup, a full configuration of all registers, and advanced features (like MLC, FSM, and pedometer, among others) embedded within digital output devices. This software visualizes the sensors' output in both numeric and graphical format, permitting the user to manage or save data flowing from the device. The software package includes firmware (.bin and .dfu files) primed to flash on the motherboard.

ST offers FSM and ML examples, enabling developers to experiment with ST scripts and data to grasp components' capabilities. (Note: ST provides implementation examples but not the dataset, so ST assumes some data science knowledge to be able to use MLC examples.)

Apart from the Unico-GUI and Professional MEMS tool evaluation approach, IMUs with MLC are also part of the BOM of two system development Kits, where a full sensor node is implemented with ST Sensors, STM32L4+ MCU and BlueNRG BLE Modules for communication: SensorTIle.box has LSM6DSOX and STEVAL-STWINKT1 has ISM330DHCX. For more information on you can refer to the dedicated pages for these example products.

Examples of Products with Inertial Sensors with Machine Learning Cores