CausalVAD：通过因果干预去混淆的端到端自动驾驶
CausalVAD: De-confounding End-to-End Autonomous Driving via Causal Intervention
提出稀疏因果干预模块（SCIS），可即插即用地嵌入端到端驾驶模型，通过原型字典对稀疏查询做后门调整，消除数据集偏差导致的因果混淆，提升规划安全性与鲁棒性。
原文依据 we introduce CausalVAD, a de-confounding training framework that leverages causal intervention.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论面向规划的端到端驾驶模型展现出巨大潜力，但它们从根本上学习的是统计相关性而非真正的因果关系。这种脆弱性导致因果混淆，即模型将数据集偏差作为捷径加以利用，严重损害其在复杂场景中的可靠性与安全性。为解决这一问题，我们提出 CausalVAD，一种利用因果干预的去混淆训练框架。其核心是稀疏因果干预方案（SCIS），这是一个轻量级、即插即用的模块，用于在神经网络中实现后门调整理论。SCIS 构建一个表示潜在驾驶上下文的原型字典，然后利用该字典对模型的稀疏向量化查询进行干预。这一步骤主动消除由混淆因子引发的虚假关联，从而从表征中剔除虚假因素以供下游任务使用。在 nuScenes 等基准上的大量实验表明，CausalVAD 实现了最优的规划精度与安全性。此外，我们的方法在针对诱发因果混淆而配置的数据偏差和噪声场景下展现出更优的鲁棒性。
英文原文摘要 1190 字符
Planning-oriented end-to-end driving models show great promise, yet they fundamentally learn statistical correlations instead of true causal relationships. This vulnerability leads to causal confusion, where models exploit dataset biases as shortcuts, critically harming their reliability and safety in complex scenarios. To address this, we introduce CausalVAD, a de-confounding training framework that leverages causal intervention. At its core, we design the sparse causal intervention scheme (SCIS), a lightweight, plug-and-play module to instantiate the backdoor adjustment theory in neural networks. SCIS constructs a dictionary of prototypes representing latent driving contexts. It then uses this dictionary to intervene on the model's sparse vectorized queries. This step actively eliminates spurious associations induced by confounders, thereby eliminating spurious factors from the representations for downstream tasks. Extensive experiments on benchmarks like nuScenes show CausalVAD achieves state-of-the-art planning accuracy and safety. Furthermore, our method demonstrates superior robustness against both data bias and noisy scenarios configured to induce causal confusion.