从近100篇综述出发，一览自动驾驶的技术发展路线！

huangguang

2023年快要过去了，不得不说，今年的技术变更实在很快，在线高精地图、大模型、端到端自动驾驶、世界模型、Occ、Nerf这些新兴技术，慢慢走向量产的计划中，今天自动驾驶Daily就为大家盘下近百篇综述和经典论文，涉及感知、定位、融合、Occupancy、大模型、端到到、规划控制、BEV感知、数据相关等，一览自动驾驶发展路线。

所有论文出自--国内首个自动驾驶学习社区：自动驾驶之心知识星球（点我有惊喜），所有论文均可在星球内下载，更有30+学习路线，近2300人一起讨论。
端到端自动驾驶

Recent Advancements in End-to-End Autonomous Driving using Deep Learning: A SurveyEnd-to-end Autonomous Driving: Challenges and Frontiers

在线高精地图

HDMapNet：基于语义分割的在线局部高精地图构建 (ICRA2022)

VectorMapNet：基于自回归方式的端到端矢量化地图构建（ICML2023）

MapTR ： 基于固定数目点的矢量化地图构建 (ICLR2023)

MapTRv2：一种在线矢量化高清地图构建的端到端框架

PivotNet：基于动态枢纽点的矢量化地图构建 （ICCV2023）

BeMapNet：基于贝塞尔曲线的矢量化地图构建 (CVPR2023)

LATR:  无显式BEV 特征的3D车道线检测 (ICCV2023)

TopoNet: 基于图的驾驶场景拓扑推理

TopoMLP: 先检测后推理（拓扑推理 strong pipeline）

LaneGAP：连续性在线车道图构建

Neural Map Prior: 神经地图先验辅助在线建图 （CVPR2023）

MapEX：现有地图先验显著提升在线建图性能

大模型与自动驾驶

CLIP：Learning Transferable Visual Models From Natural Language SupervisionBLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and GenerationBLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language ModelsInstructBLIP: Towards General-purpose Vision-Language Models with Instruction TuningMiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language ModelsInstructGPT：Training language models to follow instructions with human feedbackADAPT: Action-aware Driving Caption TransformerBEVGPT：Generative Pre-trained Large Model for Autonomous Driving Prediction, Decision-Making, and PlanningDriveGPT4：Interpretable End-to-end Autonomous Driving via Large Language ModelDrive Like a Human Rethinking Autonomous Driving with Large Language ModelsDriving with LLMs: Fusing Object-Level Vector Modality for Explainable Autonomous DrivingHiLM-D: Towards High-Resolution Understanding in Multimodal Large Language Models for Autonomous DrivingLanguageMPC: Large Language Models as Decision Makers for Autonomous DrivingPlanning-oriented Autonomous DrivingWEDGE A multi-weather autonomous driving dataset built from generative vision-language models

Nerf与自动驾驶

NeRF: Neural Radiance Field in 3D Vision, A Comprehensive ReviewNeural Volume Rendering: NeRF And BeyondMobileNeRF：移动端实时渲染，Nerf导出Mesh（CVPR2023）Co-SLAM：实时视觉定位和NeRF建图（CVPR2023）Neuralangelo：当前最好的NeRF表面重建方法（CVPR2023）MARS：首个开源自动驾驶NeRF仿真工具（CICAI2023）UniOcc：NeRF和3D占用网络（AD2023 Challenge）Unisim：自动驾驶场景的传感器模拟（CVPR2023）

Occupancy占用网络

Grid-Centric Traffic Scenario Perception for Autonomous Driving: A Comprehensive Review

BEV感知

Vision-Centric BEV Perception: A SurveyVision-RADAR fusion for Robotics BEV Detections: A SurveySurround-View Vision-based 3D Detection for Autonomous Driving: A SurveyDelving into the Devils of Bird’s-eye-view Perception: A Review, Evaluation and Recipe

多模态融合

针对Lidar、Radar、视觉等数据方案进行融合感知；

A Survey on Deep Domain Adaptation for LiDAR PerceptionAutomatic Target Recognition on Synthetic Aperture Radar Imagery：A SurveyDeep Multi-modal Object Detection and Semantic Segmentation for Autonomous Driving：Datasets, Methods, and ChallengesMmWave Radar and Vision Fusion for Object Detection in Autonomous Driving：A ReviewMulti-Modal 3D Object Detection in Autonomous Driving：A SurveyMulti-modal Sensor Fusion for Auto Driving Perception：A SurveyMulti-Sensor 3D Object Box Refinement for Autonomous DrivingMulti-View Fusion of Sensor Data for Improved Perception and Prediction in Autonomous Driving

3D检测

对基于单目图像、双目图像、点云数据、多模态数据的3D检测方法进行了梳理；

3D Object Detection for Autonomous Driving：A Review and New Outlooks3D Object Detection from Images for Autonomous Driving A SurveyA Survey of Robust LiDAR-based 3D Object Detection Methods for autonomous drivingA Survey on 3D Object Detection Methods for Autonomous Driving ApplicationsDeep Learning for 3D Point Cloud Understanding：A SurveyMulti-Modal 3D Object Detection in Autonomous Driving：a surveySurvey and Systematization of 3D Object Detection Models and Methods

目标检测综述

主要涉及通用目标检测任务、检测任务中的数据不均衡问题、伪装目标检测、自动驾驶领域检测任务、anchor-based、anchor-free、one-stage、two-stage方案等；

A Survey of Deep Learning for Low-Shot Object DetectionA Survey of Deep Learning-based Object DetectionCamouflaged Object Detection and Tracking：A SurveyDeep Learning for Generic Object Detection：A SurveyImbalance Problems in Object Detection：A surveyObject Detection in 20 Years：A SurveyObject Detection in Autonomous Vehicles：Status and Open ChallengesRecent Advances in Deep Learning for Object Detection

目标检测数据增强与不均衡问题

主要涉及目标检测任务中的数据增强、小目标检测、小样本学习、autoargument等工作；

A survey on Image Data Augmentation for Deep LearningAugmentation for small object detectionBag of Freebies for Training Object Detection Neural NetworksGeneralizing from a Few Examples：A Survey on Few-ShotLearning Data Augmentation Strategies for Object Detection

分割综述

主要对实时图像分割、视频分割、实例分割、弱监督/无监督分割、点云分割等方案展开讨论；

A Review of Point Cloud Semantic SegmentationA SURVEY ON DEEP LEARNING METHODS FOR SEMANTIC IMAGE SEGMENTATION IN REAL-TIMEA SURVEY ON DEEP LEARNING METHODS FOR SEMANTICA Survey on Deep Learning Technique for Video SegmentationA Survey on Instance Segmentation State of the artA Survey on Label-efficient Deep Segmentation-Bridging the Gap between Weak Supervision and Dense PredictionA Technical Survey and Evaluation of Traditional Point Cloud Clustering  for LiDAR Panoptic SegmentationEvolution of Image Segmentation using Deep Convolutional Neural Network A SurveyOn Efficient Real-Time Semantic SegmentationUnsupervised Domain Adaptation for Semantic Image Segmentation-a Comprehensive Survey

多任务学习

对检测+分割+关键点+车道线联合任务训练方法进行了汇总；

Cascade R-CNNDeep Multi-Task Learning for Joint Localization, Perception, and PredictionMask R-CNNMask Scoring R-CNNMulti-Task Multi-Sensor Fusion for 3D Object DetectionMultiTask-CenterNetOmniDetYOLOPYOLO-Pose

目标跟踪

对单目标和多目标跟踪、滤波和端到端方法进行了汇总；

Camouflaged Object Detection and Tracking：A SurveyDeep Learning for UAV-based Object Detection and Tracking：A SurveyDeep Learning on Monocular Object Pose Detection and Tracking：A Comprehensive OverviewDetection, Recognition, and Tracking：A SurveyInfrastructure-Based Object Detection and Tracking for Cooperative Driving Automation：A SurveyRecent Advances in Embedding Methods for Multi-Object Tracking：A SurveySingle Object Tracking：A Survey of Methods, Datasets, and Evaluation MetricsVisual Object Tracking with Discriminative Filters and Siamese Networks：A Survey and Outlook

深度估计

针对单目、双目深度估计方法进行了汇总，对户外常见问题与精度损失展开了讨论；

A Survey on Deep Learning Techniques for Stereo-based Depth EstimationDeep Learning based Monocular Depth Prediction：Datasets, Methods and ApplicationsMonocular Depth Estimation Based On Deep Learning：An OverviewMonocular Depth Estimation：A SurveyOutdoor Monocular Depth Estimation：A Research ReviewTowards Real-Time Monocular Depth Estimation for Robotics：A Survey

关键点检测

人体关键点检测方法汇总，对车辆关键点检测具有一定参考价值；

2D Human Pose Estimation：A SurveyA survey of top-down approaches for human pose estimationEfficient Annotation and Learning for 3D Hand Pose Estimation：A SurveyRecent Advances in Monocular 2D and 3D Human Pose Estimation：A Deep Learning Perspective

Transformer综述

视觉transformer、轻量级transformer方法汇总；

A Survey of Visual TransformersA Survey on Visual TransformerEfficient Transformers：A Survey

车道线检测

对2D/3D车道线检测方法进行了汇总，基于分类、检测、分割、曲线拟合等；
2D车道线

A Keypoint-based Global Association Network for Lane DetectionCLRNet：Cross Layer Refinement Network for Lane DetectionEnd-to-End Deep Learning of Lane Detection and Path Prediction for Real-Time Autonomous DrivingEnd-to-end Lane Detection through Differentiable Least-Squares FittingKeep your Eyes on the Lane：Real-time Attention-guided Lane DetectionLaneNet：Real-Time Lane Detection Networks for Autonomous DrivingTowards End-to-End Lane Detection：an Instance Segmentation ApproachUltra Fast Structure-aware Deep Lane Detection

3D车道线

3D-LaneNet+：Anchor Free Lane Detection using a Semi-Local RepresentationDeep Multi-Sensor Lane DetectionFusionLane：Multi-Sensor Fusion for Lane Marking Semantic Segmentation Using Deep Neural NetworksGen-LaneNet：A Generalized and Scalable Approach for 3D Lane DetectionONCE-3DLanes：Building Monocular 3D Lane Detection3D-LaneNet：End-to-End 3D Multiple Lane Detection

SLAM综述

定位与建图方案汇总；

A Survey on Active Simultaneous Localization and Mapping-State of the Art and New FrontiersThe Revisiting Problem in Simultaneous Localization and Mapping-A Survey on Visual Loop Closure DetectionFrom SLAM to Situational Awareness-ChallengesSimultaneous Localization and Mapping Related Datasets-A Comprehensive Survey

模型量化

A Survey on Deep Neural Network CompressionChallenges, Overview, and SolutionsPruning and Quantization for Deep Neural Network Acceleration A Survey