In search of the right app
ST's SL-AIAID012401V1 kit provides a streamlined way for companies to experiment with machine learning for automotive applications. It consists of the AEKD-AICAR1 evaluation board, an AI plugin, and integration with the AutoDevKit platform. The kit's pre-trained algorithm can recognize four states of a car: parked, normal road conditions, bumpy road, and swerving or skidding.
This first machine learning solution from ST offers a unique opportunity for testing and developing AI applications for vehicles. Many automotive companies are still evaluating whether machine learning makes strategic sense at this point. However, building an algorithm from the ground up requires significant effort and investment.
By providing a turnkey evaluation solution on AutoDevKit, developers can more easily explore the potential of this technology in automotive. The kit handles the work of algorithm creation, so users can focus on exploring the subject matter and how machine learning may or may not fit their specific needs and goals within the industry. Overall, the SL-AIAID012401V1 lowers the barrier to entry for automotive companies interested in experimenting with this emerging technology domain.
Finding the right tool
The AEKD-AICAR1 evaluation kit aims to help developers create compelling applications, whether utilizing machine learning algorithms or not. It includes an MCU with 4MB of flash memory and the AEK-LCD-DT028V1 display. This allows teams to work on user interfaces and more general purposes.
The main board further provides two CAN FD transceivers and two potentiometers for experimenting with the analog-to-digital converter. As a result, the base could connect to an actuator board to control motors, handle wireless communication, or manage LEDs among other capabilities. This versatility permits working on additional vehicle systems beyond machine learning.
However, as the product name implies, our team designed the AEKD-AICAR1 with the goal of making machine learning more approachable for the automotive sector. The kit remains flexible enough to help teams innovating across different domains, while specifically addressing the need to explore and apply emerging AI technologies within the automotive industry through its inclusion of an AI plugin and pre-trained algorithm. Overall, it lowers barriers to advancing intelligent vehicle applications.
AEKD-AICAR1: The solution to make machine learning more accessible
Guiding developers
The AEKD-AICAR1 kit includes the AEK-CON-SENSOR1 connectivity board and an AIS2DW12 3-axis accelerometer. Developers can collect data from the MEMS sensor and utilize a long short-term memory (LSTM) recurrent neural network (RNN) to determine the car's state out of four possible options. Helpfully, the evaluation board comes pre-loaded with a pre-trained neural network on the MCU, reducing setup time. This initial LSTM RNN model was developed in a Google Colab environment using TensorFlow 2.4.0 and then optimized as a C code library using our proprietary tools.
To better guide and support our community, a comprehensive user manual is provided. This covers getting started with Google Colab, model training, data acquisition and more. Users can implement one of many popular frameworks like TensorFlow to create, train and validate their own neural networks. This empowers teams to leverage existing community resources. The guide also demonstrates how to leverage our AI plugin to optimize algorithms for deployment on microcontrollers. In essence, we share our expertise and utilities to assist developers in easily testing ML applications and determining whether the technology is suitable for their automotive solutions.
Using real-world environments
At the culmination of this process, engineers obtain a program capable of facilitating experimentation using sensors, evaluating requirements, and enhancing comprehension of feasibility – all at a substantially reduced cost compared to building from the ground up. The inclusion of a connectivity board with the AEKD-AICAR1 set allows simple swapping of alternative sensors to broaden experimentation. Frequently, the stringent safety and reliability demands within automotive engineering confine developers from immediately testing ideas using off-the-shelf consumer modules. The AEKD-AICAR1 addresses this by empowering experimentation on an industrial-grade platform that numerous engineers leverage to create real-world automotive solutions. In the end, users gain crucial insights and a functional prototype to advance their project as needed.
Conceiving original applications
Given that machine learning in automotive is still in its nascent stages, users frequently seek the most versatile platforms. Indeed, leveraging a single system across multiple projects can help conserve time and costs. For instance, one customer adjusted their LSTM nodes to examine suspension responses and determine appropriate spring compensation improvements to enhance the driving experience. Similarly, developers could employ sensors to create novel applications such as anticipating a battery's state of charge or affixing sensors to the steering wheel to monitor the driver's heart rate. In essence, the adaptability of the AEKD-AICAR1 extends beyond typical vehicular use cases and state identification, allowing exploration of diverse concepts. Its flexibility means the kit can grow in usefulness as machine learning in automotive evolves.
