Scaling Learning-based AEB with Massive Unlabeled Data

2026-06-17

AI Summary

This work addresses the challenge of scaling learning-based automatic emergency braking (AEB) systems under production constraints by leveraging massive unlabeled fleet data while mitigating false activations caused by erroneous pseudo-labels. To this end, we propose a stabilized meta-feedback semi-supervised learning (MF-SSL) framework that incorporates a noise-aware decoupling mechanism to filter out ambiguous anchor samples. Additionally, we introduce kinematics-gated pseudo-labeling and a teacher–student conflict penalty strategy to suppress risk hallucinations on unlabeled data. The resulting student model, trained on billion-scale trajectory windows, has been deployed across hundreds of thousands of vehicles. Empirical validation over 10⁹ kilometers demonstrates a true-to-false activation ratio exceeding 100:1 and a 35% improvement in accident-free mileage compared to a rule-based baseline.

Abstract

This paper studies how to scale learning-based automatic emergency braking (AEB) with massive unlabeled fleet data under production constraints. Our approach is based on meta-feedback semi-supervised learning (MF-SSL), where a teacher generates pseudo labels for unlabeled driving data and is updated using a small labeled anchor set as safety-critical feedback. In production, anchor ambiguity and labeled-unlabeled mismatch can amplify systematic pseudo-label errors, leading to spurious triggers. We propose a stabilized MF-SSL framework with (i) Noise-Aware Decoupling, which removes ambiguity-prone anchors from the teacher's supervised update path, and (ii) kinematics-gated pseudo-labeling with a teacher conflict penalty to suppress mismatch-induced risk hallucinations on unlabeled data while maintaining broad coverage. Extensive experiments show consistent gains as unlabeled data scale from 1M to 1B windows, improving safety while keeping comfort stable. The 1B-trained student model is deployed to hundreds of thousands of vehicles and validated over \$10^9$ km of driving, achieving a positive-to-false activation ratio exceeding 100:1 and a 35% improvement in accident-free driving mileage over a production rule-only baseline.

Problem

• automatic emergency braking
• semi-supervised learning
• pseudo-labeling
• unlabeled data
• production constraints

Innovation

• meta-feedback semi-supervised learning
• Noise-Aware Decoupling
• kinematics-gated pseudo-labeling
• automatic emergency braking
• large-scale unlabeled data