ST's SPC58EC80E5, AIS2DW12-based AI driving status detection solution

With today's focus on improving vehicle safety, new technologies are making it easier to protect drivers and their passengers. ST' provides a good solution that detects vehicle motion and road conditions, warning the driver when the road is uneven and bumpy or when the vehicle is skidding.

AEKD-AICAR1 is a versatile deep learning system based on a long short-term memory (LSTM) recurrent neural network (RNN).AEKD-AICAR1 is built on a 4 MB flash general-purpose SPC58EC automotive microcontroller and uses an ultra-low-power AIS2DW12 three-axis motion sensor to acquire information on changes in vehicle motion and current road conditions.

The acquired data is passed to a LSTM (long short-term memory) recurrent neural network running on the MCU, which classifies driving conditions and generates an accurate predictive model of the vehicle state, which is displayed on the LCD touch screen.

To obtain a pre-trained neural network, developers can use one of the many available frameworks (Tensorflow, Lasagne, etc.) and an external IDE (Google Colab) to create, train and validate (using truth tables) an LSTM recurrent neural network network.

Key Product Benefits:

AIS2DW12 - ultra-low power 3-axis digital motion sensor;

• -Ultra-low power consumption;
• -Embedded self-test function;
• - Small, thin plastic LGA package;
• - Two supply voltage options (1.8V or 3.3V);

 SPC58EC80E5 - 32-bit MCU for automotive applications;

• -4MB of flash memory;
• -Support for the ASIL-B safety standard in accordance with ISO 26262;
• -Rich communication interfaces;
• - Embedded hardware safety module (HSM) with dedicated flash memory;

 For other information about the basic principles of neural network, Long Short Term Memory Recurrent Neural Network (LSTM RNN), and designing an artificial intelligence car sensing node, please refer to the attached manual.

►Scene application diagram

►Display board photo

►Solution Block Diagram

►Core technology advantage

• Running pre-trained neural networks on "simple" MCUs;
• Dedicated long short-term memory (LSTM) recurrent neural networks for time series analysis;
• Artificial intelligence;
• AEC-Q100 compliance;
• compact hardware, with ST's intelligent algorithms, high CP value;

►Specifications

• Artificial intelligence;
• Real-time analysis of sensor acceleration;
• Dedicated LSTM recurrent neural network for time series analysis;

Recognition of four car states:

1. Cars parked or stopped;
2. Cars driving on roads with normal conditions;
3. Cars driving on rough roads;
4. Cars skidding or swerving;