analiza aiqbench
This paper applies sparse autoencoders to the internal activations of a simple single-camera driving planner, and reports that the learned features align with route intent, speed, and braking, and respond predictably to traffic lights and stop signs. The results are on a small model, and the 3.44% gain from the adapter is modest, so the main contribution is a method for probing driving models, not a deployable safety tool. Readers working on interpretability or autonomous-driving ML should watch whether these SAE methods scale to larger end-to-end stacks.
Mechanistic Interpretability for End-to-End Self-Driving
arXiv:2610.41185v1 Announce Type: cross Abstract: End-to-end (E2E) autonomous driving models achieve strong planning performance but remain largely opaque, limiting observability and trust. We apply mechanistic interpretability to E2E driving by training Sparse Autoencoders (SAEs) on the internal activations of a simple, single-camera end-to-end path planning model. Unsupervised clustering of SAE latents reveals interpretable structure aligned with route intent, vehicle speed, and braking behavior, recovering route intent with 98.3% accuracy without supervision. Through paired counterfactual image edits, we further show that specific SAE features are visually grounded and systematically respond to scene changes such as traffic light state and stop sign presence. Steering individual SAE latents produces predictable, monotonic changes in planner outputs including lateral displacement, curvature, and braking. Finally, we demonstrate a practical application of these representations: we build a lightweight adapter trained on SAE latents that reduces trajectory selection error by 3.44%, outperforming an equivalent adapter that omits SAE latents, confirming that the sparse representation captures useful information for interpretability. Code and visualizations are available at https://mgagvani.github.io/mechanistic-interpretability-for-e2e-driving/
Abstract:End-to-end (E2E) autonomous driving models achieve strong planning performance but remain largely opaque, limiting observability and trust. We apply mechanistic interpretability to E2E driving by training Sparse Autoencoders (SAEs) on the internal activations of a simple, single-camera end-to-end path planning model. Unsupervised clustering of SAE latents reveals interpretable structure aligned with route intent, vehicle speed, and braking behavior, recovering route intent with 98.3% accuracy without supervision. Through paired counterfactual image edits, we further show that specific SAE features are visually grounded and systematically respond to scene changes such as traffic light state and stop sign presence. Steering individual SAE latents produces predictable, monotonic changes in planner outputs including lateral displacement, curvature, and braking. Finally, we demonstrate a practical application of these representations: we build a lightweight adapter trained on SAE latents that reduces trajectory selection error by 3.44%, outperforming an equivalent adapter that omits SAE latents, confirming that the sparse representation captures useful information for interpretability. Code and visualizations are available at this https URL
Accepted to the Conference on Robot Learning (CoRL) 2026. Project page: this https URL
Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2610.41185 [cs.RO]
(or arXiv:2610.41185v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2610.41185 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 281 cuvinte
Subiecte conexe
Beyond Static Interpretability: Anticipating Post-SFT Mechanisms from Pre-SFT Parameters for Better Tuning
arXiv · 27 aug. 2026
Mechanistic Interpretability of Structure-Aware Numerical Reasoning in LLaMA 3.1 8B
arXiv · 21 aug. 2026
Global Challenges in AI Safety for Biosecurity
LessWrong · 20 sept. 2026
Introducing PIRAMID: Physics-Informed Research for Ambitious Mechanistic Interpretability
LessWrong · 25 iul. 2026