ECCV 2026 | 自动驾驶相关论文汇总(端到端/世界模型/OCC等)

ECCV 放榜也有一段时间了。自动驾驶之心汇总了近期公开的部分论文，后续有计划分析下当下的自动驾驶研究方向。欢迎大家加入自动驾驶之心知识星球持续关注~

一、端到端自动驾驶

BEVLM: Distilling Semantic Knowledge from LLMs into Bird's-Eye View Representations

文章链接：https://arxiv.org/abs/2603.06576

Deconfounded Lifelong Learning for Autonomous Driving via Dynamic Knowledge Spaces

文章链接：https://arxiv.org/abs/2603.14354

开源链接：https://github.com/Mooncakebro/DeLL

ExploreVLA: Dense World Modeling and Exploration for End-to-End Autonomous Driving

文章链接：https://arxiv.org/abs/2604.02714

开源链接：https://zihaosheng.github.io/ExploreVLA/

PixelPilot: Scalable Vision-Language-Action Models for End-to-End Autonomous Driving

文章链接：https://arxiv.org/abs/2607.04637

PriorEye: Geospatial Visual Priors for End-to-End Autonomous Driving

文章链接：https://arxiv.org/abs/2606.31830

开源链接：https://ori-mrg.github.io/PriorEye

Teaching Vision-Language-Action Models What to See and Where to Look

文章链接：https://arxiv.org/abs/2607.01658

开源链接：https://github.com/ShivaTeam/DriveTeach-VLA

UniTeD: Unified Temporal Diffusion for Joint Perception and Planning in Autonomous Driving

文章链接：https://arxiv.org/abs/2606.25736

二、世界模型与驾驶生成

Driver-WM: A Driver-Centric Traffic-Conditioned Latent World Model for In-Cabin Dynamics Rollout

文章链接：https://arxiv.org/abs/2605.05092

DriveVA: Video Action Models are Zero-Shot Drivers

文章链接：https://arxiv.org/abs/2604.04198

LaGen: Towards Autoregressive LiDAR Scene Generation

文章链接：https://arxiv.org/abs/2511.21256

开源链接：https://github.com/szzhou88/LaGen

Learning Transferable Dynamics Priors from Action to World Modeling

文章链接：https://arxiv.org/abs/2606.29501

Long-term Traffic Simulation via Structured Autoregressive Modeling

文章链接：https://arxiv.org/abs/2606.31209

NavWM: A Unified Navigation World Model for Foresight-Driven Planning

文章链接：https://arxiv.org/abs/2606.24101

OmniNWM: Omniscient Driving Navigation World Models

文章链接：https://arxiv.org/abs/2510.18313

开源链接：https://arlo0o.github.io/OmniNWM/

UniDrive-WM: Unified Understanding, Planning and Generation World Model for Autonomous Driving

文章链接：https://arxiv.org/abs/2601.04453

开源链接：https://unidrive-wm.github.io/UniDrive-WM

三、轨迹预测与运动规划

Towards Metric-Agnostic Trajectory Forecasting

文章链接：https://arxiv.org/abs/2607.01133

Unveiling Transferability in Trajectory Prediction via Latent Scene Embeddings

文章链接：https://arxiv.org/abs/2606.30777

四、多传感器融合

Horizon3D: Sparse Radar-Camera Fusion for Long-Range 3D Perception in Autonomous Driving

文章链接：https://arxiv.org/abs/2606.31096

Open-Weather Robust 3D Detection via Dual-Critic Diffusion Alignment

文章链接：https://arxiv.org/abs/2607.01983

RAF: Reliability-Aware Fusion of Camera, LiDAR, and 4D RADAR for Robust 3D Object Detection in Adverse Weather

文章链接：https://arxiv.org/abs/2607.04587

开源链接：https://github.com/parkie0517/RAF

RESOLVE: A Multi-Resolution and Multi-Modal Dataset for Roadside Cooperative Perception

文章链接：https://arxiv.org/abs/2606.31895

开源链接：https://github.com/ASU-Suo-Lab/RESOLVE

五、占用网格与3D场景理解

Cross-View Yaw Estimation in Location Uncertainty with Line-Aligning Yaw Scoring

文章链接：https://arxiv.org/abs/2606.22094

FDR-Occ: Factorized Dense Routing for Full-Spectrum 3D Occupancy Prediction

文章链接：https://arxiv.org/abs/2607.03822

FLM-Occ: Feed-forward Likelihood Maximization for Efficient Indoor Occupancy Prediction

文章链接：https://arxiv.org/abs/2606.21373

Generative Lane Topology Reasoning via Autoregressive Model with Geometry Prior

文章链接：https://arxiv.org/abs/2606.31814

HilDA: Hierarchical Distillation with Diffusion for Advancing Self-Supervised LiDAR Pre-training

文章链接：https://arxiv.org/abs/2606.20189

开源链接：https://maxiuw.github.io/hilda

HSDF-Lane: Height-Aligned Signed Distance Field with Semantic Lane Prior for 3D Lane Detection

文章链接：https://arxiv.org/abs/2606.31172

Open-Vocabulary BEV Segmentation with 3D-Aware Geometric Constraints

文章链接：https://arxiv.org/abs/2606.24353

开源链接：https://hchoi256.github.io/projects/ovbevseg/

RoadBench: Benchmarking MLLMs on Fine-Grained Spatial Understanding and Reasoning under Urban Road Scenarios

文章链接：https://arxiv.org/abs/2511.18011

Streaming Dense Voxel Representations for 3D Occupancy Prediction

文章链接：https://arxiv.org/abs/2503.22087

开源链接：https://moonseokha.github.io/StreamOcc/

Think While You Map: Asynchronous Vision-Language Agents for Incremental 3D Scene Graphs

文章链接：https://arxiv.org/abs/2606.31471

开源链接：https://denizbickici.github.io/thinkgraphs/

六、目标检测跟踪与分割

Comprehensive Robustness Analysis of LiDAR-based 3D Object Detection in Autonomous Driving

文章链接：https://arxiv.org/abs/2607.02074

LeAD-M3D: Leveraging Asymmetric Distillation for Real-Time Monocular 3D Detection

文章链接：https://arxiv.org/abs/2512.05663

PLOT: Pseudo-Labeling via Object Tracking for Monocular 3D Object Detection

文章链接：https://arxiv.org/abs/2507.02393

开源链接：https://plot-eccv.github.io

七、VLM/LLM 赋能驾驶

CritiqueDriveVLM: From Verifier-Guided Reinforcement Learning to Latent Thought Distillation for Autonomous Driving

文章链接：https://arxiv.org/abs/2607.04179

开源链接：https://github.com/MICLAB-BUPT/CritiqueDriveVLM

EgoDyn-Bench: Evaluating Ego-Motion Understanding in Vision-Centric Foundation Models for Autonomous Driving

文章链接：https://arxiv.org/abs/2604.22851

开源链接：https://tum-avs.github.io/EgoDyn-Bench-Website/

八、仿真、测试与安全

CCFM: Collision-Constrained Flow Matching for Safety-Critical Scenario Generation

文章链接：https://arxiv.org/abs/2607.04451

开源链接：https://github.com/KELISBU/CCFM

DriveWeaver: Point-Conditioned Video Inpainting for Controllable Vehicle Insertion in Autonomous Driving Simulation

文章链接：https://arxiv.org/abs/2606.31918

FrozenDrive: Zero-Shot Text-Guided Driving Scene Generation and Data Augmentation with Parameter-Free Frozen Diffusion Model

文章链接：https://arxiv.org/abs/2606.20110

Understanding Cross-Rig Generalization in Automotive Perception: a Multi-Rig Benchmark and Rig Variation Metrics

文章链接：https://arxiv.org/abs/2606.27554

九、其他

ASTAD: Asymmetric Style Transfer for Synthetic-to-Real Adaptation in Autonomous Driving

文章链接：https://arxiv.org/abs/2606.29286

开源链接：https://github.com/Dingyi-Yao/ASTAD

MapDreamer: Aerial Imagery Conditioned Latent Diffusion for Lane-Level Map Generation

文章链接：https://arxiv.org/abs/2607.01370

国内首个自动驾驶全栈技术社区

『自动驾驶之心知识星球』成立三年多啦~我们是一个技术+产业+求职，三位一体的自驾社区，实现了多个领域的闭环。我们致力于为行业培养更多优秀的人才，提供更多机会。新人加入通道已经开启，欢迎新同学扫码加入！

不是搬运信息，只过滤认知

每天各种自驾资讯刷屏。论文、融资、技术路线……信息越多，越不知道信谁。

我们星球的核心价值不是“发得多”，而是“选得准”。这里有身处学术前沿的嘉宾，有 Tier 1行业研究员这样的产业观察者，有从业者本人分享真机踩坑经验。

同样的论文，你看摘要，他们看代码可行性；同样的融资新闻，你看热闹，他们看技术路线走向。

技术+产业+求职闭环

我们发现很多社群单纯地讨论炼丹、代码、复现，是技术的乌托邦，但自驾的行业趋势一直在变，我们意识到单纯的技术分享很难满足大家的实际需求。

在星球里，你会看到：技术+产业+求职三者形成闭环。

技术拆解：VLA、世界模型、端到端最新论文深度解读

产业洞察：头部公司技术路线分析、量产落地进展跟踪、创业方向、投融资

求职内推： Tier 1公司人才结构，岗位实时更新，简历直推业务负责人

不是“单向输出”，是“双向互动”

这不是一个“星主发帖、成员围观”的地方。

可以向嘉宾提问，可以在评论区争论，可以在社群中找到志同道合的同行。这里的活跃度，不是靠“签到”撑起来的，是靠真实的行业讨论撑起来的。我们会不定期和一线的学术界&工业界大佬畅聊自动驾驶发展趋势，探讨技术走向和量产痛点：

加入星球有哪些福利？

• 第一时间掌握自动驾驶相关的学术进展、工业落地应用；
• 和行业大佬一起交流工作与求职相关的问题；
• 优良的学习交流环境，能结识更多同行业的伙伴；
• 星球内部专属学习视频，搭配文档不枯燥；
• 自动驾驶相关工作岗位推荐，第一时间对接企业；
• 行业机会挖掘，投资与项目对接。