This talk examines the core architectural choices in autonomous driving: modular, model-based ADAS versus end-to-end deep learning. Modular systems offer interpretability and compatibility with safety standards like ISO 26262, while end-to-end approaches promise greater adaptability through data-driven learning. However, the latter challenges current validation and certification frameworks. We explore technical trade-offs, real-world deployment implications, and how hybrid architectures may bridge the gap between performance and trustworthiness.
In this session, you will learn more about:
- The technical and safety-related advantages and limitations of modular model-based ADAS vs. end-to-end learning architectures
- How architectural decisions affect explainability, traceability, and verifiability under automotive safety standards
- Hybrid system designs that blend modular interpretability with data-driven adaptability for robust autonomous performance
&nsbp;
powered by Borlabs Cookie