Auditing Latent-Space Monitors for Autonomous Driving

2026-09-24

AI Summary

This study investigates whether accessing internal model representations genuinely outperforms relying solely on external inputs and outputs for fault monitoring in autonomous driving. To this end, we audit latent space monitors for the LaneSegNet and VAD models, proposing the first post-hoc frame-level fault monitoring protocol for online vectorized map generation. By employing supervised latent probes and multimodal annotations, we systematically compare the predictive efficacy of internal versus external features. Our findings reveal that internal representations provide no statistically significant gain for fault prediction; notably, leveraging output data alone achieves superior accuracy, yielding a VAD AUROC of 0.924. These results demonstrate that internal access is unnecessary for effective fault monitoring. Furthermore, we release an open-source frame-level fault label dataset to facilitate future research in this domain.

Abstract

Runtime failure monitors can use a model's internal representations to anticipate failures. We audit this monitoring strategy across two autonomous-driving tasks: online vectorized map generation with LaneSegNet and end-to-end planning with VAD. We find that frame-level errors are predictable at inference in both tasks. For LaneSegNet, a supervised latent probe reaches Area Under the Receiver Operating Characteristic curve (AUROC) 0.780 for high Chamfer error; to our knowledge, this is the first post-hoc frame-level failure monitor for online vectorized map generation. For VAD, a supervised planning-latent probe reaches AUROC 0.868 for mean-ADE failure. Our audit shows that internal access is not necessary for strong failure prediction. A monitor using only LaneSegNet's prediction outputs reaches AUROC 0.825, while for VAD, ego state, driving command, and the planner's predicted trajectory reach 0.924 on the same mean-ADE endpoint. Adding latent features to either baseline yields no statistically resolved improvement. This observation persists across a broad suite of planning failure endpoints, including endpoints whose labels depend on geometry unavailable to the non-latent baseline. Thus, predicting failure from an internal representation does not establish that the representation provides useful information beyond observable inputs and outputs. We propose an evaluation protocol for testing the incremental value of latent access and release our per-frame failure endpoint labels.

Problem

• latent-space monitors
• autonomous driving
• runtime failure prediction
• internal representations

Innovation

• Latent-Space Monitors
• Autonomous Driving
• Failure Prediction
• Evaluation Protocol
• Runtime Monitoring