xiaofan0100/Elevator-LIO-Dataset

Elevator-LIO Dataset

面向电梯非惯性运动和跨楼层定位的 LiDAR-惯性里程计数据集。包含 20 条自采真实场景序列，涵盖 79 次电梯乘坐，场景包括办公楼、宿舍、校园和商场等环境。下载的主数据文件均为 ROS 1 bag。

A LiDAR-inertial odometry dataset for elevator-induced non-inertial motion and multi-floor localization. It contains 20 self-collected real-world sequences with 79 elevator rides across office, dormitory, campus, and mall environments. The released main data files are ROS 1 bags.

---

社区贡献 / Community Contributions

我们欢迎社区贡献者分享自己的电梯相关 rosbag。社区贡献数据位于 community_contributions/，不会被转换脚本的 --all 选项自动选中；可以把相对路径作为参数单独转换。

We welcome community contributors to share elevator-related rosbag files. Community data is stored under community_contributions/. It is not selected automatically by the converter's --all option, but each relative bag path can be converted explicitly.

Item	Description
Storage path	community_contributions/
Contact	xiaofan@sjtu.edu.cn

Contributor	Platform / ID
@编程猫小渐	小红书 9556244270

---

数据概览 / Dataset Overview

序列 Sequence	文件 Files	场景 Scene
Campus 1-4	Campus*.bag	校园 / Campus
Dormitory 1-4	Dormitory*.bag	宿舍楼 / Dormitory
Mall 1-2	Mall*.bag	商场 / Shopping Mall
Office 1-10	Office*.bag	办公楼 / Office Building

覆盖场景：大规模跨楼层建图、长距离垂直位移、电梯轿厢内手持晃动、动态行人和镜面内饰。

Covered scenarios: large-scale multi-floor mapping, long vertical travel, handheld motion in elevator cabins, dynamic pedestrians, and mirrored interiors.

---

传感器配置 / Sensor Setup

Component	Description
LiDAR	Livox MID-360 with built-in IMU
Camera	Synchronized industrial camera (Mall2.bag and Office3.bag)
Compute	Jetson Orin Nano
Data format	ROS 1 bag; optional conversion to ROS 2 Humble rosbag2

ROS 话题与消息类型 / ROS Topics and Message Types

所有主序列均包含：

All main sequences contain:

Topic	ROS 1 source type	ROS 2 converted type
/livox/lidar	livox_ros_driver2/CustomMsg	livox_ros_driver2/msg/CustomMsg
/livox/imu	sensor_msgs/Imu	sensor_msgs/msg/Imu

Mall2.bag 和 Office3.bag 还包含 /camera/image/compressed、/camera/exposure_time 和 /camera/gain。部分序列包含 /LIO/set_elevator_flag（std_msgs/Bool）。这些标准消息在转换后使用对应的 ROS 2 pkg/msg/Type 名称。

Mall2.bag and Office3.bag additionally contain /camera/image/compressed, /camera/exposure_time, and /camera/gain. Some sequences contain /LIO/set_elevator_flag (std_msgs/Bool). These standard messages use their corresponding ROS 2 pkg/msg/Type names after conversion.

---

ROS 1 bag 转换为 ROS 2 / Convert ROS 1 Bags to ROS 2

数据集根目录提供 convert_rosbag1_to_rosbag2.py。它将 ROS 1 bag 转为 ROS 2 Humble 可用的 SQLite3 rosbag2 目录，并将 Livox 点云转码为 CDR；输出的 LiDAR 类型名称和字段布局与真实 MID-360 的 livox_ros_driver2/msg/CustomMsg 一致。转换过程本身不要求 source ROS 环境。

The dataset root provides convert_rosbag1_to_rosbag2.py. It converts ROS 1 bags into ROS 2 Humble-compatible SQLite3 rosbag2 directories and transcodes Livox packets to CDR. The output LiDAR type name and field layout match the real MID-360 driver's livox_ros_driver2/msg/CustomMsg. No sourced ROS environment is required for conversion.

安装 Python 依赖 / Install the Python dependency:

python3 -m pip install --user rosbags

转换一个或多个序列 / Convert one or more sequences:

cd Elevator-LIO-Dataset
python3 convert_rosbag1_to_rosbag2.py Office1
python3 convert_rosbag1_to_rosbag2.py Campus1 Dormitory1

转换全部 20 个主序列 / Convert all 20 main sequences:

python3 convert_rosbag1_to_rosbag2.py --all

默认输出到同级的 Elevator-LIO-Dataset-rosbag2/。也可以指定容量更大的磁盘，并先用 --dry-run 检查话题、消息数量和空间需求：

The default output root is the sibling directory Elevator-LIO-Dataset-rosbag2/. You can select a larger disk and use --dry-run first to inspect topics, message counts, and disk requirements:

python3 convert_rosbag1_to_rosbag2.py --all --dry-run
python3 convert_rosbag1_to_rosbag2.py --all --output-dir /data/Elevator-LIO-Dataset-rosbag2

脚本内置官方 Livox ROS Driver 2 的消息字段作为回退；如果检测到驱动源码，它会先验证字段一致性。也可显式指定驱动消息目录：

The script embeds the official Livox ROS Driver 2 message fields as a fallback. If a driver checkout is detected, its schema is verified first. A driver message directory can also be selected explicitly:

python3 convert_rosbag1_to_rosbag2.py Office1 \
  --livox-msg-dir ~/ws_livox_ros_driver2/src/livox_ros_driver2/msg

每个输出完成前，脚本会验证各话题消息数量，并解码每种类型的首条 CDR 消息。播放时需要先 source ROS 2、Humble 版 Livox 驱动和 Elevator-LIO 工作空间：

Before publishing each output directory, the script verifies topic counts and decodes the first CDR message of every type. Source ROS 2, the Humble Livox driver, and Elevator-LIO before playback:

source /opt/ros/humble/setup.bash
source <livox_ros_driver2_ws>/install/setup.bash
source <elevator_lio_ros2_ws>/install/setup.bash
ros2 bag info ../Elevator-LIO-Dataset-rosbag2/Office1
ros2 bag play ../Elevator-LIO-Dataset-rosbag2/Office1

---

内外参 / Calibration

标定文件为 calibration_offsets.yaml。变换约定如下：

The calibration file is calibration_offsets.yaml. The transform convention is:

p_target = target_R_source * p_source + target_t_source

Field	Meaning	Direction
imu_R_lidar, imu_t_lidar	LiDAR pose in IMU frame / IMU 系下激光雷达位姿	LiDAR to IMU
lidar_R_vehicle, lidar_t_vehicle	Vehicle center pose in LiDAR frame / LiDAR 系下小车中心位姿	Vehicle to LiDAR
cam_R_lidar, cam_t_lidar	LiDAR pose in camera frame / Camera 系下激光雷达位姿	LiDAR to Camera
camera.K	Camera intrinsic matrix / 相机内参矩阵	Camera model
camera.dist	OpenCV coefficients k1, k2, p1, p2, k3 / 畸变参数	Camera model

---

文件结构 / File Structure

.
├── Campus1.bag ... Campus4.bag
├── Dormitory1.bag ... Dormitory4.bag
├── Mall1.bag, Mall2.bag
├── Office1.bag ... Office10.bag
├── community_contributions/       # 社区贡献数据 / community data
├── calibration_offsets.yaml       # 内外参 / calibration
├── check_bag_files.py             # 完整性校验 / integrity check
├── convert_rosbag1_to_rosbag2.py  # ROS 1 -> ROS 2 converter
└── README.md

快速校验原始 ROS 1 bag / Validate the original ROS 1 bags:

python3 check_bag_files.py .

---

代码仓库 / Code Repository

算法代码 / Source code: github.com/xiaofan4122/Elevator-LIO

项目主页 / Project page: xiaofan4122.github.io/Elevator_LIO_Page

---

许可证 / License

本数据集采用 CC BY 4.0 许可。可自由分享和改编，包括商业用途，需注明出处。

Licensed under CC BY 4.0. Free to share and adapt for any purpose, including commercial use, with attribution.

---

引用 / Citation

@article{zhang2026elevatorlio,
  title={Elevator-LIO: Robust LiDAR-Inertial Odometry for Multi-Floor Navigation under Elevator-Induced Non-Inertial Motion},
  author={Zhang, Yifan and Huang, Yudong and Zhang, Yuchong and Li, Changze and Liu, Haoran and Yang, Ming and Qin, Tong},
  journal={arXiv preprint arXiv:2605.24495},
  year={2026}
}