STMicroelectronics' first vehicle gauge machine learning solution, SL-AIAID012401V1, consists of the AEKD-AICAR1 evaluation Kit, AIAI plug-in and AutoDevKit Vehicle gauge development board. It can recognize four car states: parking, normal road conditions, rough roads, wheel sidescarding or sudden swerving.
This is a rare opportunity to test and develop automotive AI applications to determine if the technology is suitable for this market. In fact, many car companies are still exploring whether machine learning makes sense for them at this stage of the industry. It takes a lot of people and money to create an algorithm from scratch. Bringing the evaluation solution to our AutoDevKit platform makes it easier for developers to explore the feasibility of AI for vehicle regulation.
The overall architecture of vehicle status recognition system
While machine learning is steadily gaining traction in the consumer market, the automotive industry is still exploring what machine learning technology can bring to the car. Artificial intelligence has been a hot topic in the automotive industry for decades, and researchers have been exploring how to apply machine learning in navigation systems since as early as 2001. However, AI tends to be used in several application scenarios, such as autonomous driving or accident detection.
In addition, for safety and reliability reasons, the automotive industry has relied on heuristics for decades. In short, car manufacturers are already good at delivering advanced features without having to train neural networks. As a result, development platforms must lower the technical barriers to automotive machine learning without discouraging designers from envisioning more traditional applications.
Find the right tools
To help developers create their killer app, whether it uses machine learning algorithms or not, the AEKD-AICAR1 comes with an MCU with 4mb of flash memory, as well as an AEK-LCD-DT028V1 display for UIs and more general applications. The motherboard also offers two CAN FD transceivers and two potentiometers for testing the performance of the analog-to-digital converter. As a result, this circuit board can be connected to an actuator board to drive motors, handle wireless communication functions, or manage devices such as leds. As a result, the board is flexible enough to help development teams develop applications with other systems. However, as the name suggests, our development team developed applications with AEKD-AICAR1 because the automotive industry demanded a lower technical threshold for machine learning.
AEKD-AICAR1 makes machine learning better for use in solutions
AEKD-AICAR1 package
The AEKD-AICAR1 is equipped with an AEK-CON-SENSOR1 connector board and an AIS2DW12 three-axis accelerometer. Developers can take MEMS data and use long short-term memory (LSTM) recurrent neural networks (RNNS) to determine the car's four states. The kit is also designed to simplify the development of demo applications, as the AEKD-AICAR1 comes pre-trained neural networks inside the MCU flash memory. The current version of the LSTM RNN is a neural network developed in the Google Colab environment using the TensorFlow 2.4.0 framework, and then we use ST tools to convert the pre-trained neural network into a C code base suitable for running on the STM32.
To further guide and train developers, we have provided users with a user's guide to the Google Colab environment, model training, data acquisition and other tools. Users can create, train, and validate neural networks using many popular machine learning frameworks such as TensorFlow, enabling teams to borrow from many existing community projects. We then showed how the AI plug-in used by ST's neural network optimizes the algorithm to run on a microcontroller. In short, we share our knowledge experiences and utilities so that developers can more easily test projects and determine if machine learning is suitable for automotive applications.
Use real environment
Eventually, engineers will get a program that allows them to try to develop new applications with sensors, assess their needs, and better understand what is feasible within a fraction of the cost of starting from scratch. The connector board is included in the AEKD-AICAR1 bundle, which means users can easily switch to another sensor for further evaluation. The safety and reliability limitations inherent in the automotive industry prevent developers from testing automotive product concepts with commercially available consumer-grade boards. The AEKD-AICAR1 solves this problem by experimenting on an automotion-grade platform that many people use to develop practical applications.
Imagine original applications
Since machine learning in the automotive industry is still in its infancy, users often look for platforms with the greatest flexibility. In fact, developing multiple projects with a single system can help save time and resources, for example, one customer made a small change to the LSTM node to analyze the response performance of the vehicle's suspension and determine the degree of spring compensation, thereby improving the driving experience. Similarly, developers can use sensors to develop new applications, for example, to predict the state of a battery's charge, or to install sensors on the steering wheel to check the driver's heart rate. In short, the AEKD-AICAR1 is flexible and not limited to traditional automotive applications and condition detection.
Heisener Electronic is a famous international One Stop Purchasing Service Provider of Electronic Components. Based on the concept of Customer-orientation and Innovation, a good process control system, professional management team, advanced inventory management technology, we can provide one-stop electronic component supporting services that Heisener is the preferred partner for all the enterprises and research institutions.
