<ul><li>Autonomous systems utilizing Machine Learning (ML) face challenges in ensuring reliability and safety due to lack of human-understandable explanations for errors in ML predictions.</li><li>A new fuzzy-based monitoring approach has been introduced in this paper, specifically designed for ML perception components to provide explanations on the impact of various conditions on perception reliability and to act as a safety monitor during operation.</li><li>The effectiveness of the proposed monitor was evaluated using naturalistic driving datasets in an automated driving scenario, highlighting the interpretability of the monitor and identifying reliable performance under certain operating conditions.</li><li>Benchmarking showed that the novel monitor improved safety by avoiding hazardous situations while maintaining mission performance, outperforming existing runtime ML monitors in the assessed dataset.</li></ul>

Explaining Unreliable Perception in Automated Driving: A Fuzzy-based Monitoring Approach

Discover more