In real-world driving, resolving a primary hazard often precipitates a secondary one: swerving to avoid a stalled vehicle may force the ego car into oncoming traffic. Existing driving QA datasets fail to capture this cascading causality, evaluating perception, prediction, and planning in silos without anticipating the risks action induces. We introduce WaymoQA, a multi-view dataset that frames safety-critical reasoning as a two-stage task: resolving the immediate threat, then mitigating the action-induced secondary risk. WaymoQA provides 35,000 human-annotated QA pairs across image and video modalities, with 28K training and 6.4K test examples. Experiments show that SOTA models struggle on safety-critical scenes, while fine-tuning on WaymoQA improves both reasoning accuracy and consistency. WaymoQA thus serves as a rigorous benchmark and resource for safety-aware autonomous driving.

Vision-Language Models (VLMs) have shown remarkable performance on diverse visual and linguistic tasks, yet they remain fundamentally limited in their understanding of 3D spatial structures.We propose Geometric Distillation, a lightweight, annotation-free fine-tuning framework that injects human-inspired geometric cues into pretrained VLMs without modifying their architecture.By distilling (1) sparse correspondences, (2) relative depth relations, and (3) dense cost volumes from off-the-shelf 3D foundation models (e.g., MASt3R, VGGT), our method shapes representations to be geometry-aware while remaining compatible with natural image–text inputs.Through extensive evaluations on 3D vision-language reasoning and 3D perception benchmarks, our method consistently outperforms prior approaches, achieving improved 3D spatial reasoning with significantly lower computational cost.Our work demonstrates a scalable and efficient path to bridge 2D-trained VLMs with 3D understanding, opening up wider use in spatially grounded multimodal tasks.