九月SLAM相关论文速递

九月SLAM相关论文速递

论文列表
DirectTracker: 3D Multi-Object Tracking Using Direct Image Alignment and Photometric Bundle Adjustment
3D VSG: Long-term Semantic Scene Change Prediction through 3D Variable Scene Graphs
Leveraging Large Language Models for Robot 3D Scene Understanding
D-Lite: Navigation-Oriented Compression of 3D Scene Graphs under Communication Constraints
WiSion: Bolstering MAV 3D Indoor State Estimation by Embracing Multipath of WiFi
General Place Recognition Survey: Towards the Real-world Autonomy Age
ExplORB-SLAM: Active Visual SLAM Exploiting the Pose-graph Topology
Continual Adaptation of Semantic Segmentation using Complementary 2D-3D Data Representations
Differential Flatness-Based Trajectory Planning for Autonomous Vehicles
Hilti-Oxford Dataset: A Millimetre-Accurate Benchmark for Simultaneous Localization and Mapping
E-NeRF: Neural Radiance Fields from a Moving Event Camera
Robust Incremental Smoothing and Mapping (riSAM)
Semi-Direct Point-Line Visual Inertial Odometry for MAVs
3D Reconstruction by Pretrained Features and Visual-Inertial Odometry
A Coupled Visual and Inertial Measurement Units Method for Locating and Mapping in Coal Mine Tunnel
Deep Pose Graph-Matching-Based Loop Closure Detection for Semantic Visual SLAM
Efficient 3D Lidar Odometry Based on Planar Patches
Orbeez-SLAM: A Real-time Monocular Visual SLAM with ORB Features and NeRF-realized Mapping
Efficient and Consistent Two Key-Frame Visual-Inertial Navigation Using Matrix Lie Groups

DirectTracker: 3D Multi-Object Tracking Using Direct Image Alignment and Photometric Bundle Adjustment

融合工作
SLAM得到的地图和位姿数据 + 2D分割/掩膜分割结果 + 运动预测跟踪等 = 3D目标跟踪
是一种比较新奇的方法，相较于端到端的更具有可解释性。

3D VSG: Long-term Semantic Scene Change Prediction through 3D Variable Scene Graphs

针对 语义SLAM + 长期SLAM任务中物体位置移动，进行的改进。

1. 提出三维可变图对语义场景进行建模
2. 利用网络学习得到每种物体位置变化的概率
   

Leveraging Large Language Models for Robot 3D Scene Understanding

室内家居环境中的场景理解问题。
利用语言模型进行推断，输入“一个房间包含xxx、xxx，那么这个房间是____”。输出对房间类型的推断。

D-Lite: Navigation-Oriented Compression of 3D Scene Graphs under Communication Constraints

语义地图的数据压缩问题。

只保留图中最短路径的边，从而对模型进行压缩

WiSion: Bolstering MAV 3D Indoor State Estimation by Embracing Multipath of WiFi

通过室内WIFI + IMU 进行无人机定位（经典吃瓜吃到自己家，场地用的是自己本科团队的）
实际上还是一个松耦合的融合，通过wifi和IMU分别计算位姿然后进行融合，IMU会飘、但是WIFI精度低，融合得到较好结果。

General Place Recognition Survey: Towards the Real-world Autonomy Age

场景识别综述，分为不同传感器对场景识别方法进行了概述
综述能发在T-Ro上的团队应该是很厉害的团队了。

ExplORB-SLAM: Active Visual SLAM Exploiting the Pose-graph Topology

主动SLAM,利用ORB-SLAM提供的位姿图信息和稀疏地图信息，构建一个决策方法，从而使得机器人能够主动的探索整个空间构建地图。

Hilti-Oxford Dataset: A Millimetre-Accurate Benchmark for Simultaneous Localization and Mapping

具有挑战性环境下的新数据集，位姿真实值精度达到mm级。

E-NeRF: Neural Radiance Fields from a Moving Event Camera

事件相机和NeRF结合，在高动态环境下重建

Robust Incremental Smoothing and Mapping (riSAM)

iSAM 的升级版

Semi-Direct Point-Line Visual Inertial Odometry for MAVs

融合怪类型的论文，在PL-VIO中添加，处理非关键帧就采用直接法去对齐而不提取特征，从而得到更高效的结果。

3D Reconstruction by Pretrained Features and Visual-Inertial Odometry

A Coupled Visual and Inertial Measurement Units Method for Locating and Mapping in Coal Mine Tunnel

PL-VIO 应用在隧道中的场景下

Deep Pose Graph-Matching-Based Loop Closure Detection for Semantic Visual SLAM

回环检测的一种很新颖的思路，将关键帧组成的位姿图进行匹配，回环检测变成了特征匹配问题去解决。

Efficient 3D Lidar Odometry Based on Planar Patches

ICP激光点云匹配的替代方案，利用平面特征中的共面约束

1. 能够更快更好的完成任务
2. 对场景是否有要求？
   

Orbeez-SLAM: A Real-time Monocular Visual SLAM with ORB Features and NeRF-realized Mapping

在ORB-SLAM基础上，添加NerF稠密地图构建模块。
利用了成熟ORB-SLAM框架以及算法提供的深度信息加速收敛过程，比单纯采用nerf的slam算法快几百倍。

1. 速度快，能够在i7+3090下实时运行
2. 精度高，由于采用传统方法估计的精度
3. 收敛快，利用传统方法提供的深度信息加速收敛
4. 是首个实时单目稠密地图的方法
   

Efficient and Consistent Two Key-Frame Visual-Inertial Navigation Using Matrix Lie Groups