今日从 arXiv 订阅中筛选 10 篇论文。

Arxiv ID2609.32322幻觉翻译2609.32322

⚡ Not All Errors Matter: Decision-Relevant Prediction Error Predicts Planning Quality

决策相关误差（DRPE）：世界模型的误差该落在哪，才和规划有关。

Arxiv ID2609.34440幻觉翻译2609.34440

⚡ When the Score Becomes the Target: Rethinking Metric Validity in Autonomous Driving

当分数变成优化目标：AD 指标有效性何时失效。

When the Score Becomes the Target: Rethinking Metric Validity in Autonomous Driving
Arxiv ID2609.34414幻觉翻译2609.34414

⚡ From World Models to World Action Models: Rethinking Next-State Prediction

CF-WAM：next-state 表示不该在训练前被固定。

From World Models to World Action Models: Rethinking Next-State Prediction
Arxiv ID2609.34085幻觉翻译2609.34085

⚡ AD-E2E-JEPA: A Joint-Embedding Predictive Architecture For End-to-End Autonomous Driving

JEPA 世界模型打通端到端驾驶：先体检、再提速 100×。

AD-E2E-JEPA: A Joint-Embedding Predictive Architecture For End-to-End Autonomous Driving
Arxiv ID2609.33030幻觉翻译2609.33030

⚡ What Must a World Model Distinguish for Planning?

规划只要求世界模型保住「决策充分」的区分，不要求全保真。

What Must a World Model Distinguish for Planning?
Arxiv ID2609.34563幻觉翻译2609.34563

⚡ Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence

潜空间视觉推理的证据-信用缺口：诊断清楚再修。

Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence
Arxiv ID2609.31631幻觉翻译2609.31631

⚡ OMP-MoE: Efficient Expert Pruning for Mixture-of-Experts LLMs via Orthogonal Matching Pursuit

专家剪枝 = 稀疏信号重建：OMP 选原子，water-filling 跨层分配。

OMP-MoE: Efficient Expert Pruning for Mixture-of-Experts LLMs via Orthogonal Matching Pursuit
Arxiv ID2609.34684幻觉翻译2609.34684

⚡ Natural State-Prediction Accuracy can Hide Weak Controlled Responsiveness in VLA Readouts

VLA 读出审计：能解码 ≠ 有响应。

Natural State-Prediction Accuracy can Hide Weak Controlled Responsiveness in VLA Readouts
Arxiv ID2609.32811幻觉翻译2609.32811

⚡ Progressive Risk Estimation for Accident Anticipation

事故预判：从「会不会撞」到「何时撞」的连续风险曲线。

Progressive Risk Estimation for Accident Anticipation
Arxiv ID2609.32182幻觉翻译2609.32182

⚡ KeyRec: Bounded Visual Memory for Streaming and Long-Video Understanding

流式长视频的有界视觉记忆：写读分离、事件银行、固定预算路由。

KeyRec: Bounded Visual Memory for Streaming and Long-Video Understanding

自动生成于 2026-09-30 · 基于 arXiv Daily Digest