Towards Trustworthy and Explainable AI for Perception Models: From Concept to Prototype Vehicle Deployment

Researchers built a trustworthy perception module for autonomous driving that explains its 3D scene understanding through attention-based, faithfulness-tested XAI.

The system adds uncertainty estimation and calibration, plus training methods to improve robustness in perception models.

It was demonstrated in a prototype vehicle with a real-time XAI interface, showing that explainable monitoring can work on-road.

The work supports safer deployment and oversight of autonomous driving AI by making predictions more transparent and better calibrated.