Explainable deep learning improves human mental models of self-driving cars

Concept-Wrapper Network, a concept-based explanation approach deployed on a real autonomous vehicle, improves drivers’ ability to understand and predict its behaviour. Self-driving cars increasingly rely on deep neural networks to achieve human-like driving 1 , 2 , 3 . The opacity of these black-box planners makes it challenging to accurately anticipate when they will fail 4 , 5 , 6 , with potentially catastrophic consequences 7 , 8 , 9 [1].

Sources

1. nature.com · Explainable deep learning improves human mental models of self-driving cars Sep 2, 2026

This is the neutral brief the press writes once, for everyone, before each reader’s paper adds its own so-what. It links out; it does not republish. 1 article clustered.

Print your own paper