Giters

arnoldfychen / Awesome-Autonomous-Driving

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Awesome-Autonomous-Driving Awesome

Author: 牛肉咖喱饭(PeterJaq)

Update：2023/12/03

This project will be periodically updated with quality projects and papers related to autonomous driving.

Now I am re-orging this project.

Update

  • [2023/12/3] Add Daily ADAS Arxiv Paper List!! in ADAS-Arxiv-Daily
  • [2023/12/2] Update Arxiv 2023 11 Monthly ADAS Paper List!! Arxiv-202311
  • [2023/11/20] Update NeurIPS 2023 ADAS Paper List!! NeurIPS2023
  • [2023/09/27] Update ICCV 2023 ADAS Paper List！！

Contents

1. Autonomous Driving Midleware and Integrated Solutions

1.1 Midelware

中间件

  • ROS - A set of software libraries and tools that help you build robot applications.
  • ROS-2 - A set of software libraries and tools that help you build robot applications.
  • Cyber - High performance runtime framework designed specifically for autonomous driving (AD) scenarios from baidu.

1.2 Integrated Solutions

解决方案

2. Sensor and Calibration Tools

2.1 Sensor Hardware

传感器硬件

LiDAR

Camera

GPS/IMU

MCU

2.2 Calibration Tools

参数标定工具

  • OpenCalib - ALL in One 商汤开源的自动驾驶多传感器的一个开源标定工具箱，基本涵盖了大部分的自动驾驶标定场景。
  • camera-calibration - 能够比较好的阐述相机标定具体步骤和原理的
  • CameraCalibration - 这个项目集合了相机标定相关的多个脚本工具，便于完成完整的车载环视相机标定流程
  • ros-camera-lidar-calibration - 相机内参标定与相机lidar外参标定
  • lidar_IMU_calib - Lidar IMU 的标定工具
  • sync_gps_lidar_imu_cam - lidar-imu-cam-GPS时间戳硬件同步方案

3. Perception

3.1 Detection

检测与分割

3.1.1 Vision based

基于视觉

BackBone

  • Next-ViT 来自字节面向工业界的新一代Transform模型部署。
  • CoAtNet
  • FocalsConv Focal Sparse Convolutional
  • PoolFormer [CVPR2022] MetaFormer Is Actually What You Need for Vision. 证明Transformer模型的能力，而不是设计复杂的token mixer来实现SOTA性能
  • ConvNext [CVPR2022] A ConvNet for the 2020s. 用设计transformer的**构建卷积。
  • Mobile-Former [CVPR2022] 微软提出Mobile-Former，MobileNet和Transformer的并行设计，可以实现局部和全局特征的双向融合，在分类和下游任务中，性能远超MobileNetV3等轻量级网络！
  • Up to 31 Revisiting Large Kernel Design in CNNs. 大Kernel =? SOTA 这篇文章给你答案！

Occupancy

  • Occupancy Networks Learning 3D Reconstruction in Function Space.
  • Pyramid Occupancy Network Predicting Semantic Map Representations from Images using Pyramid Occupancy Networks.
  • MonoScene Monocular 3D Semantic Scene Completion.
  • OccDepth A Depth-Aware Method for 3D Semantic Scene Completion.
  • VoxFormer Sparse Voxel Transformer for Camera-based 3D Semantic Scene.
  • TPVFormer Tri-Perspective View for Vision-Based 3D Semantic Occupancy Prediction.
  • SurroundOcc Multi-Camera 3D Occupancy Prediction for Autonomous Driving.
  • A Comprehensive Review of Occupancy A summary of the current research trend and provide some probable future outlooks for occupancy.

数据增强

  • TeachAugment [CVPR2022] Data Augmentation Optimization Using Teacher Knowledge
  • AlignMixup [CVPR2022] Improving Representations By Interpolating Aligned Features
  • rising 基于pytorch的GPU数据预处理transform模块，实测好用！

Lane Detection

Object Detection

  • YOLOR - 提出了在网络模型中引入隐知识的概念，将隐知识和显知识同时作用于模型训练，通过核函数对齐，预测精修以及多任务同时学习，让网络表征出一种统一化的特征。
  • YOLOX - Anchor-free 版本的YOLO，堆砌了解耦头，simOTA等，达到了SOTA
  • 3D-BoundingBox
  • Pseudo_Lidar_V2 - Accurate Depth for 3D Object Detection in Autonomous Driving.
  • Pseudo_lidar - Pseudo-LiDAR from Visual Depth Estimation: Bridging the Gap in 3D Object Detection for Autonomous Driving.
  • https://arxiv.org/abs/2203.10981 MonoDTR: Monocular 3D Object Detection with Depth-Aware Transformer 基于单目的Depth-Aware Transformer 的3D检测.
  • BoxeR Box-Attention for 2D and 3D Transformers. 从鸟瞰平面生成判别信息，用于 3D 端到端对象检测。该项目同样也提出了2D上的Detection 解决方案。

3.1.2 Lidar based

基于激光雷达

Object Detection

  • Voxelnet
  • Complex-YOLO
  • PointRCNN
  • CenterPoint - 3D Object Detection and Tracking using center points in the bird-eye view.
  • PartA2-Net - From Points to Parts: 3D Object Detection from Point Cloud with Part-aware and Part-aggregation Network.
  • CIA-SSD - Confident IoU-Aware Single Stage Object Detector From Point Cloud.
  • 3DIoUMatch-PVRCNN - 3DIoUMatch: Leveraging IoU Prediction for Semi-Supervised 3D Object Detection.
  • SFA3D - Super Fast and Accurate 3D Object Detection based on 3D LiDAR Point Clouds.
  • Auto4D - Auto4D: Learning to Label 4D Objects from Sequential Point Clouds.
  • 3DAL - Offboard 3D Object Detection from Point Cloud Sequences
  • LIFT [CVPR2022] LIFT: Learning 4D LiDAR Image Fusion Transformer for 3D Object Detection
  • FSD [CVPR2022] Fully Sparse 3D Object Detection & SST: Single-stride Sparse Transformer 来自图森的 Sparse Transformer.
  • VoxelNext [CVPR2023] VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking
  • PillarNext [CVPR2023] Rethinking Network Designs for 3D Object Detection in LiDAR Point Clouds
  • LargeKernel3D [CVPR2023] LargeKernel3D: Scaling Up Kernels in 3D Sparse CNNs
  • LinK [CVPR2023] Linear Kernel for LiDAR-Based 3D Perception
  • Spherical Transformer [CVPR2023]spherical Transformer for LiDAR-Based 3D Recognition
  • Unspervised 3D OD [CVPR2023]Towards Unsupervised Object Detection From LiDAR Point Clouds
  • Benchmarking robustness of 3D OD [CVPR2023] Benchmarking Robustness of 3D Object Detection to Common Corruptions
  • Bi3D [CVPR2023] Bi-Domain Active Learning for Cross-Domain 3D Object Detection
  • Density-Insensitive [CVPR2023] Density-Insensitive Unsupervised Domain Adaption on 3D Object Detection
  • UniDistill [CVPR2023] UniDistill: A Universal Cross-Modality Knowledge Distillation Framework for 3D Object Detection in Bird’s-Eye View
  • MSF [CVPR2023] MSF: Motion-Guided Sequential Fusion for Efficient 3D Object Detection From Point Cloud Sequences
  • OcTr [CVPR2023] OcTr: Octree-Based Transformer for 3D Object Detection
  • SlowLiDAR [CVPR2023] Increasing the Latency of LiDAR-Based Detection Using Adversarial Examples
  • Uni3D [CVPR2023] Uni3D: A Unified Baseline for Multi-Dataset 3D Object Detection
  • DetZero [ICCV2023] Rethinking Offboard 3D Object Detection with Long-term Sequential Point Clouds
  • FocalFormer3D [ICCV2023] Focusing on Hard Instance for 3D Object Detection
  • GPA-3D [ICCV2023] Geometry-aware Prototype Alignment for Unsupervised Domain Adaptive 3D Object Detection from Point Clouds
  • KECOR [ICCV2023] KECOR: Kernel Coding Rate Maximization for Active 3D Object Detection
  • Once Detected, Never Lost [ICCV2023] Once Detected, Never Lost: Surpassing Human Performance in Offline LiDAR based 3D Object Detection
  • PARTNER [ICCV2023] PARTNER: Level up the Polar Representation for LiDAR 3D Object Detection
  • PG-RCNN [ICCV2023] PG-RCNN: Semantic Surface Point Generation for 3D Object Detection
  • Domain-Adaptive [ICCV2023]Revisiting Domain-Adaptive 3D Object Detection by Reliable, Diverse and Class-balanced Pseudo-Labeling

Lidar Ground Segmentation

  • patchwork Patchwork 主要由三部分组成：基于同心带模型（CZM）的极坐标网格表示、区域地平面拟合（R-GPF）和地面似然估计（GLE) IROS2021
  • patchwork++ 与Patchwork不同，Patchwork++由称为反射噪声去除（RNR）、区域垂直平面拟合（R-VPF）、自适应GLE（A-GLE）和空间地面恢复（TGR）的新模块组成。Patchwork++具有更高的精确度和召回率。此外，新的Patchwork++具有较低的召回标准差。
  • TRAVEL 他使用三维点云的图形表示，同时进行可穿越的地面检测和物体聚类, 为了分割可穿越的地面，点云被编码为一个图结构，即三网格场，它将每个三网格视为一个节点。IROS 2022

Lidar Segmentation

  • RangeView [ICCV2023] Rethinking Range View Representation for LiDAR Segmentation

3.1.2 Multi Sensor Fusion

3D Object Detection

3.2 Tracking

追踪算法

3.3 High Performance Inference

高性能推理

视觉系列

  • Lite.ai - 该项目提供了一系列轻量级的目标检测语义分割任务的整合框架支持 YOLOX🔥, YoloR🔥, YoloV5, YoloV4, DeepLabV3, ArcFace, CosFace, RetinaFace, SSD, etc.
  • multi-attention -> onnx -
    提供了一个多头注意力机制支持onnx部署的方式
  • TRT ViT 字节跳动提出的面向工业界部署的ViT

LiDAR Pillars系列

4. Prediction

  • [An Auto-tuning Framework for Autonomous Vehicles] (https://arxiv.org/pdf/1808.04913.pdf)
  • VectorNet - 来自VectorNet: Encoding HD Maps and Agent Dynamics from Vectorized Representation利用高精地图 -与目标物信息进对目标进行行为预测。apollo在7.0版本的行为预测部分的encoder利用了这个vectornet.
  • TNT - TNT是一种基于历史数据（即多代理和环境之间交互）生成目标的轨迹状态序列方法，并基于似然估计得到紧凑的轨迹预测集。
  • DESIRE - DESIRE: Distant Future Prediction in Dynamic Scenes with Interacting Agents
  • TNT: Target-driveN Trajectory Prediction apollo在7.0版本的行为预测模块inter-TNT的轨迹生成利用了TNT的方法.
  • MultiPath++ - Efficient Information Fusion and Trajectory Aggregation for Behavior Prediction.
  • MotionCNN - A Strong Baseline for Motion Prediction in Autonomous Driving.
  • WAT - Weakly-supervised Action Transition Learning for Stochastic Human Motion Prediction.
  • BEVerse - Unified Perception and Prediction in Birds-Eye-View for Vision-Centric Autonomous Driving.
  • ParkPredict+ - Vehicle simualtion and behavior prediction in parking lots.
  • HiVT - Hierarchical Vector Transformer for Multi-Agent Motion Prediction
  • FEND A Future Enhanced Distribution-Aware Contrastive Learning Framework for Long-Tail Trajectory Prediction
  • EqMotion Equivariant Multi-Agent Motion Prediction With Invariant Interaction Reasoning
  • EigenTrajectory [ICCV2023] EigenTrajectory: Low-Rank Descriptors for Multi-Modal Trajectory Forecasting
  • Temporal Enhanced [ICCV2023] Temporal Enhanced Training of Multi-view 3D Object Detector via Historical Object Prediction
  • TrajectoryFormer [ICCV2023] TrajectoryFormer: 3D Object Tracking Transformer with Predictive Trajectory Hypotheses

5 Localization and SLAM

Localization

  • hdl_localization - Lidar + IMU 基于卡尔曼滤波的位置估计使用了激光雷达，IMU, 可以做到实时估计。

SLAM

  • PaGO-LOAM 一个基于LeGO-LOAM的LiDAR测距框架，在这个框架中，测试地面分割算法是否有助于提取特征和改善SLAM性能是很容易和直接的。
  • Quatro-LeGO-LOAM 在城市环境中避免退化的鲁棒性global registration方法 + LeGO-LOAM
  • AVP-SLAM来自2020IROS的AVP定位方案：AVP-SLAM: Semantic Visual Mapping and Localization for Autonomous Vehicles in the Parking Lot(IROS 2020),主要是通过BEV视角对停车场中的车道线车库线以及标识进行检测并利用其进行稀疏定位。 最近有两位大佬提供了仿真和定位的开源方案：AVP-SLAM-SIM AVP-SLAM-PLUS
  • DeepLIO - Lidar + IMU 一款基于深度学习的lidar IMU融合里程计
  • hdl_graph_slam - Lidar + IMU + GPS 它基于三维图形SLAM，具有基于NDT扫描匹配的测距估计和循环检测。它还支持几个约束，如GPS、IMU。
  • LIO-SAM - Lidar + IMU + GPS 基于激光雷达，IMU和GPS多种传感器的因子图优化方案，以及在帧图匹配中使用帧-局部地图取代帧-全局地图。
  • LVI-SAM - Lidar + Camera 基于视觉+激光雷达的惯导融合
  • LeGO-LOAM - Lidar LeGO-LOAM是以LOAM为框架而衍生出来的新的框架。其与LOAM相比，更改了特征点的提取形式，添加了后端优化，因此，构建出来的地图就更加的完善。
  • SC-LeGO-LOAM - Lidar LeGO-LOAM的基于全局描述子Scan Context的回环检测
  • SC-LIO-SAM - Lidar + Camera LIO-SAM的基于全局描述子Scan Context的回环检测
  • Livox-Mapping - **Livox + IMU + SC ** 一款基于Livox的mapping工具包，在先前的工具上添加了SC和Fastlio的一些特性
  • Fast-LIO - 目前最好用的前端里程计之一，几乎同时兼具速度与鲁棒性
  • Faster-LIO - 比Fast LIO快1-1.5倍的前端里程计
  • FAST_LIO_SLAM Scancontext + 现在的SOTA里程计(fast lio)
  • SC-A-LOAM - Scancontext + 现在的SOTA里程计(Lego-loam, fast lio, a loam etc.)
  • FAST_LIO_LOCALIZATION Fast lio 系列建图完成后依赖这些执行定位.
  • Deep Functional Maps Understanding and Improving Features Learned in Deep Functional Maps
  • vMap vMAP: Vectorised Object Mapping for Neural Field SLAM
  • DeepLSD DeepLSD: Line Segment Detection and Refinement with Deep Image Gradients
  • EgoLoc [ICCV2023] EgoLoc: Revisiting 3D Object Localization from Egocentric Videos with Visual Queries

6. Planning

规划

  • 自动驾驶中的决策规划算法概述
  • 有限状态机
  • MPC
  • PathPlanning
  • pacmod - Designed to allow the user to control a vehicle with the PACMod drive-by-wire system.
  • rrt - C++ RRT (Rapidly-exploring Random Tree) implementation.
  • HypridAStarTrailer - A path planning algorithm based on Hybrid A* for trailer truck.
  • path_planner - Hybrid A* Path Planner for the KTH Research Concept Vehicle.
  • fastrack - A ROS implementation of Fast and Safe Tracking (FaSTrack).
  • commonroad - Composable benchmarks for motion planning on roads.
  • traffic-editor - A graphical editor for robot traffic flows.
  • steering_functions - Contains a C++ library that implements steering functions for car-like robots with limited turning radius.
  • moveit - Easy-to-use robotics manipulation platform for developing applications, evaluating designs, and building integrated products.
  • flexible-collision-library - A library for performing three types of proximity queries on a pair of geometric models composed of triangles.
  • aikido - Artificial Intelligence for Kinematics, Dynamics, and Optimization.
  • casADi - A symbolic framework for numeric optimization implementing automatic differentiation in forward and reverse modes on sparse matrix-valued computational graphs.
  • ACADO Toolkit - A software environment and algorithm collection for automatic control and dynamic optimization.
  • CrowdNav - Crowd-aware Robot Navigation with Attention-based Deep Reinforcement Learning.
  • ompl - Consists of many state-of-the-art sampling-based motion planning algorithms.
  • openrave - Open Robotics Automation Virtual Environment: An environment for testing, developing, and deploying robotics motion planning algorithms.
  • teb_local_planner - An optimal trajectory planner considering distinctive topologies for mobile robots based on Timed-Elastic-Bands.
  • pinocchio - A fast and flexible implementation of Rigid Body Dynamics algorithms and their analytical derivatives.
  • rmf_core - The rmf_core packages provide the centralized functions of the Robotics Middleware Framework (RMF).
  • global_racetrajectory_optimization - This repository contains multiple approaches for generating global racetrajectories.
  • toppra - A library for computing the time-optimal path parametrization for robots subject to kinematic and dynamic constraints.
  • tinyspline - TinySpline is a small, yet powerful library for interpolating, transforming, and querying arbitrary NURBS, B-Splines, and Bézier curves.
  • dual quaternions ros - ROS python package for dual quaternion SLERP.
  • mb planner - Aerial vehicle planner for tight spaces. Used in DARPA SubT Challenge.
  • ilqr - Iterative Linear Quadratic Regulator with auto-differentiatiable dynamics models.
  • EGO-Planner - A lightweight gradient-based local planner without ESDF construction, which significantly reduces computation time compared to some state-of-the-art methods.
  • pykep - A scientific library providing basic tools for research in interplanetary trajectory design.
  • am_traj - Alternating Minimization Based Trajectory Generation for Quadrotor Aggressive Flight.
  • GraphBasedLocalTrajectoryPlanner - Was used on a real race vehicle during the Roborace Season Alpha and achieved speeds above 200km/h.
  • Ruckig - Instantaneous Motion Generation. Real-time. Jerk-constrained. Time-optimal.

7. Control

控制

  • PID
  • Open Source Car Control - An assemblage of software and hardware designs that enable computer control of modern cars in order to facilitate the development of autonomous vehicle technology.
  • control-toolbox - An efficient C++ library for control, estimation, optimization and motion planning in robotics.
  • mpcc - Model Predictive Contouring Controller for Autonomous Racing.
  • open_street_map - ROS packages for working with Open Street Map geographic information.
  • autogenu-jupyter - This project provides the continuation/GMRES method (C/GMRES method) based solvers for nonlinear model predictive control (NMPC) and an automatic code generator for NMPC.
  • OpEn - A solver for Fast & Accurate Embedded Optimization for next-generation Robotics and Autonomous Systems.

9. Dataset and Competition

数据集与竞赛

  • KITTI
  • BDD100k
  • UrbanNav - 一个在亚洲城市峡谷（包括东京和香港）收集的开源本地化数据集,主要用于解决定位算法的各种问题。
  • ONCE
  • SODA10M
  • OPV2V - 首个大型自动驾驶协同感知数据集 + banchmark代码框架, 由UCLA提供

10. Data Loop & Model Loop

数据闭环

NAS

  • Beta-DARTS Beta-Decay Regularization for Differentiable Architecture Search
  • ISNAS-DIP Image-Specific Neural Architecture Search for Deep Image Prior

主动学习

Coner case & Long-tail

  • RAC Retrieval Augmented Classification for Long-Tail Visual Recognition*

数据挖掘

  • AirDet Few-Shot Detection without Fine-tuning for Autonomous Exploration. 这篇文章把他放在数据挖掘方面是思考有没有可能用极少样本不用fine-tuning 后可以从原有自动驾驶数据湖中挖掘出更多的样本。

Data Requirement

OOD

11. Visualization

可视化工具

  • Carla-birdeye-view - 可以对接carla的自动驾驶鸟瞰图组件。
  • Uber AVS - 自动驾驶可视化前端组件 xviz 与 streetscape.gl
  • Cruise - Cruise 开源的一款自动驾驶前端可视化套件

12. Simulation

  • UniSim [CVPR2023] A Neural Closed-Loop Sensor Simulator
  • LiDar-in-the-loop [CVPR2023] LiDAR-in-the-Loop Hyperparameter Optimization
  • Compact Representation [CVPR2023] Learning Compact Representations for LiDAR Completion and Generation
  • MixSim [CVPR2023] MixSim: A Hierarchical Framework for Mixed Reality Traffic Simulation
  • The Differentiable Lens [CVPR2023] Compound Lens Search Over Glass Surfaces and Materials for Object Detection

13. Others

其他更好的分享

About

License:Apache License 2.0