Beyond the Beep: Scalable Collision Anticipation and Real-Time Explainability with BADAS-2.0

TL;DR

BADAS-2.0 advances collision anticipation by expanding datasets, enabling real-time edge deployment through knowledge distillation, and providing explainability with attention heatmaps and reasoning models.

Contribution

It introduces a scalable, accurate, and explainable collision anticipation system with a large dataset, edge-compatible models, and real-time interpretability features.

Findings

Expanded dataset from 40k to 178,500 videos with high-risk scenarios.

Achieved 7-12x speedup with near-parity accuracy in compact models.

Produced real-time attention heatmaps and textual explanations for predictions.

Abstract

We present BADAS-2.0, the second generation of our collision anticipation system, building on BADAS-1.0, which showed that fine-tuning V-JEPA2 on large-scale ego-centric dashcam data outperforms both academic baselines and production ADAS systems. BADAS-2.0 advances the state of the art along three axes. (i) Long-tail benchmark and accuracy: We introduce a 10-group long-tail benchmark targeting rare and safety-critical scenarios. To construct it, BADAS-1.0 is used as an active oracle to score millions of unlabeled drives and surface high-risk candidates for annotation. Combined with Nexar's Atlas platform for targeted data collection, this expands the dataset from 40k to 178,500 labeled videos (~2M clips), yielding consistent gains across all subgroups, with the largest improvements on the hardest long-tail cases. (ii) Knowledge distillation to edge: Domain-specific self-supervised pre-training on 2.25M unlabeled driving videos enables distillation into compact models, BADAS-2.0-Flash (86M) and BADAS-2.0-Flash-Lite (22M), achieving 7-12x speedup with near-parity accuracy, enabling real-time edge deployment. (iii) Explainability: BADAS-2.0 produces real-time object-centric attention heatmaps that localize the evidence behind predictions. BADAS-Reason extends this with a vision-language model that consumes the last frame and heatmap to generate driver actions and structured textual reasoning. Inference code and evaluation benchmarks are publicly available.