NXP and COMPREDICT are teaming up to accelerate Edge AI in automotive applications through software-defined sensing.

Our partner COMPREDICT's Virtual Sensors replace costly hardware sensors with AI models that deliver real-time, scalable and continuously improving estimations.

NXP's S32 processors and eIQ Auto SDK provide the hardware and software foundation for efficient deployment, including model preparation, optimization and compilation for constrained devices.

Learn more about our demos and partnership opportunities on the COMPREDICT NXP Partner Profile Page.

Highlights

Starting from a trained model - originally in TensorFlow but compatible with other popular frameworks like PyTorch, MATLAB and various open-source options - we first compiled it with FP32 precision for the target ARM Cortex-R52 cores
Initial execution took place on Synopsys' virtual development kit, enabling pre-silicon validation
Using a dataset provided by COMPREDICT, the model was then quantized to INT8, significantly reducing its footprint while balancing inference accuracy
Leveraging NXP's eIQ Auto compiler and the S32K5 virtual development kit, we successfully deployed the quantized model on the R52 core, demonstrating efficient edge AI execution on a zonal microcontroller - precisely where this use case is intended for production

Resources

MoreLess

Automotive Edge AI: Virtual Sensors

Highlights

Resources