New auditing strategy for autonomous driving failure monitors unveiled

Researchers have developed a new auditing strategy for runtime failure monitors in autonomous driving systems. The strategy was tested on two tasks: vectorized map generation using LaneSegNet and end-to-end planning with VAD. Findings indicate that frame-level errors are predictable during inference for both tasks. For LaneSegNet, a latent probe achieved an AUROC of 0.780 for high Chamfer error, and for VAD, a planning-latent probe reached an AUROC of 0.868 for mean-ADE failure. Notably, the study found that internal model representations did not offer significant improvements over monitors using only observable inputs and outputs, suggesting that latent access may not be necessary for effective failure prediction. AI

IMPACT This research could lead to more robust and interpretable failure detection systems in autonomous vehicles, potentially improving safety and trust.

RANK_REASON The cluster contains an academic paper detailing a new research methodology and findings in AI safety for autonomous driving. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• LaneSegNet
• Nikhil Kamalkumar Advani
• ScienceCast
• VAD

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →