PopcornSAR provides support for migrating AI models to Adaptive Applications (AAs) while considering integration with AI-specific open-source frameworks such as TensorFlow and PyTorch.

When migrating complex AI models implemented with TensorFlow to AAs, issues can arise. Some problems include lower compiler versions in AI SDKs that incorporate PyTorch modules provided by semiconductor companies, lack of support for certain functionalities required for Adaptive AUTOSAR, and mismatches with the essential C++ version for Adaptive AUTOSAR, resulting in compilation and operational issues for AAs.

To address these challenges, PopcornSAR modifies its Adaptive AUTOSAR Stack to align with the compiler version used by AI. We provide a proprietary SOME/IP module implemented in Python to enable integration between AI implemented in Python and control-oriented AAs. Additionally, PopcornSAR offers support to develop Python-based AI into C++ based Adaptive AUTOSAR AI once AI model validation is complete, integrating the Adaptive AUTOSAR Stack into existing AI SDKs.

Furthermore, we also provide an ARM-based Docker container virtual ECU that operates on the ARM development board and development PC without modifying the logic of AI models implemented in Python.

※ Examples of PopcornSAR's solutions and AI integration: NXP's S32G2 and eIQ-AUTO (AI SDK)