Kauã Kuhl completed Machine Learning for Automotive Safety

Built a perception validation dossier end to end for a synthetic Aurora-7 detector: defined IoU-based detection correctly, computed aggregate precision/recall and VRU false-negative rate, then sliced recall by lighting, visibility, and class rarity to expose the night/occlusion failure modes. Used those results to write a conditional release verdict grounded in the weakest slice and tied the analysis back to the Uber Tempe crash as a functional-insufficiency case.

• Leakage-Free Detection Definition
• Safety-Critical Metric Selection
• Per-Condition Failure Analysis
• Evidence-Based Release Gating