As the automotive industry transitions toward increasingly autonomous capabilities, machine learning (ML) is becoming foundational in enabling perception, decision-making and control functions across ADAS and automated driving (AD) systems. However, the inherent uncertainty and non-determinism of ML models present unique safety challenges, particularly in environments where traditional rule-based validation is insufficient. To address this, the concept of trained supervision is emerging as a critical layer in the development and deployment of safe ML-powered systems.

• Safe ML definition
• Embedded systems and SDV
• Trained supervision