tc-datasets

为自动驾驶导航提供的真实世界数据集

A real-world dataset provided for autonomous driving navigation

原始信息汇总

Tsukuba Challenge 2021 Course

TC2021, MapIV, SLAM Map Data

• Short Name: map_tc21_mapiv
• Provider (Team): MapIV
• Type: Map
• Location: Tsukuba Challenge 2021 Course
• File: 0.15_map_all.pcd (505 MB), map_converted*.pcd (10.9 GB)
• Size: 505 MB, 10.9GB
• Format: pcd
• Number of Points: 31,559,485, 682,377,762
• Point Type:
  • XYZ: Yes
  • Intensity: Yes
  • Color: No
  • Normal: No
• SLAM Method: MapIV Engine (HESAI PandarXT-32 + Septentrio mosaic)
• Description: The LiDAR measurement of more than 70 meters is cut off. "map_converted*.pcd" are raw point cloud maps, and "0.15_map_all.pcd" is a downsampled and concatenated map of them. We used Voxel Grid Filter for downsamapling. Map coordinate system is the Japan Plane Rectangular CS IXj, and its height is orthometric height.
• License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)

Tsukuba Challenge 2019 Course

TC2019, fuRo, GNSS+INS Map Data

• Short Name: map_tc19_gnss+ins_furo
• Provider (Team): fuRo
• Type: Map
• Location: Tsukuba Challenge 2019 Course
• File: map_tc19_gnss+ins_furo.pcd
• Size: 1.95 GB
• Format: pcd
• Number of Points: 83,178,268
• Point Type:
  • XYZ: Yes
  • Intensity: Yes
  • Color: No
  • Normal: No
• SLAM Method: No SLAM (GNSS+INS, using NovAtel SPAN-CPT7 + Velodyne VLP-16)
• Description: The lidar measurement of more than 100 meters is cut off. We ran the Tsukuba Challenge 2019 course, excluding the forest in the park. Latitude, longitude, and ellipsoid height of the origin: 36.08254144, 140.07642281, 66.9479.
• License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)

TC2019, fuRo, Sensor Data

• Short Name: tc19_furo
• Provider (Team): fuRo
• Type: Sensor
• Location: Tsukuba Challenge 2019 Course
• File: tc19_js_2019-09-14-14-12-15.bag (56 GB), tc19_js_2019-09-14-14-12-15.bag.7z (22 GB)
• Size: 55.8 GB
• Format: rosbag
• Date: 2019-09-14 14:12:15
• Duration: 1hr 47:31s
• Setup: Differential Wheeled Robot (Joystick Operation)
• Sensors:
  • Lidar: SureStar R-Fans-16M
  • Camera: No
  • Radar: No
  • GNSS: No
  • IMU: Xsens MTi-3
  • Motor Encoders (Wheel Odometry): Yes
• Description: Low cost 3D-Lidar. Low accuracy wheel odometry.
• License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)

TC2019, fuRo, Map Data

• Short Name: map_tc19_furo
• Provider (Team): fuRo
• Type: Map
• Location: Tsukuba Challenge 2019 Course
• File: map_tc19_o085_f-04_t05.pcd
• Size: 683 MB
• Format: pcd
• Number of Points: 22,356,688
• Point Type:
  • XYZ: Yes
  • Intensity: Yes
  • Color: No
  • Normal: Yes
• SLAM Method: Occupancy Voxel Mapping using 3D Cartographer
• Description: Moving objects have been removed.
• License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)

TC2019, Tsuchiura Project, Sensor Data

• Short Name: tc19_tsuchiura-pj
• Provider (Team): Tsuchiura Project
• Type: Sensor
• Location: Tsukuba Challenge 2019 Course
• File: 2019-11-10-13-37-16.bag
• Size: 12.8 GB
• Format: rosbag
• Date: 2019-11-10 13:37:16
• Duration: 53:18s
• Setup: Mobile Robot (Autonomous Operation)
• Sensors:
  • Lidar: Hokuyo YVT-X002, UTM-30LX-EW, URM-40LC-EW
  • Camera: Ricoh Theta S, Logicool C920
  • Radar: No
  • GNSS: u-blox NEO-M8T
  • IMU: No
  • Motor Encoders (Wheel Odometry): Yes
• Description: This bag file is compressed with 7z.
• License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)

Tsukuba Challenge 2018 Course

TC2018, fuRo, Sensor Data

• Short Name: tc18_furo
• Provider (Team): fuRo
• Type: Sensor
• Location: Tsukuba Challenge 2018 Course
• File: tc18_js_2018-09-15-14-31-29.bag (41 GB), tc18_js_2018-09-15-14-31-29.bag.7z (12 GB)
• Size: 40.8 GB
• Format: rosbag
• Date: 2018-09-15 14:31:29 (Converted to rosbag on 2018-09-20)
• Duration: 1hr 31:43s
• Setup: Differential Wheeled Robot (Joystick Operation)
• Sensors:
  • Lidar: Velodyne VLP-16
  • Camera: No
  • Radar: No
  • GNSS: No
  • IMU: Xsens MTi-3
  • Motor Encoders (Wheel Odometry): Yes
• Description: High accuracy wheel odometry.
• License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)

TC2018, fuRo, Map Data

• Short Name: map_tc18_furo
• Provider (Team): fuRo
• Type: Map
• Location: Tsukuba Challenge 2018 Course
• File: map_tc18_o085_f-04_t30.pcd
• Size: 519 MB
• Format: pcd
• Number of Points: 17,002,094
• Point Type:
  • XYZ: Yes
  • Intensity: Yes
  • Color: No
  • Normal: Yes
• SLAM Method: Occupancy Voxel Mapping using 3D Cartographer
• Description: Moving objects have been removed.
• License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)

Other Courses

iias Tsukuba 2023, Aqua (University of Tsukuba), Sensor Data

• Short Name: iias_tsukuba_2023
• Provider (Team): Aqua (University of Tsukuba)
• Type: Sensor
• Location: iias Tsukuba 2023
• File: iias_tsukuba_2023-03-02-01-06-59.bag
• Size: 16.7 GB
• Format: rosbag
• Date: 2023-03-02 10:06:59
• Duration: 30:19s
• Setup: Mobile Robot (Joystick Operation)
• Sensors:
  • Lidar: Velodyne VLP-16, Hokuyo UTM-30LX
  • Camera: No
  • Radar: No
  • GNSS: No
  • IMU: No
  • Motor Encoders (Wheel Odometry): Yes
• Description:
• License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)

UTsukuba 2022, University of Tsukuba, Sensor Data

• Short Name: utsukuba22_university_of_tsukuba
• Provider (Team): University of Tsukuba
• Type: Sensor
• Location: Tsukuba campus 2022
• File: utsukuba22_university_of_tsukuba.bag.zst
• Size: 8 GB
• Format: rosbag
• Date: 2022-07-09 10:34:13
• Duration: 47:02s
• Setup: Mobile Robot (Joystick Operation)
• Sensors:
  • Lidar: Velodyne VLP-16
  • Camera: No
  • Radar: No
  • GNSS: No
  • IMU: LOAD MicroStrain 3DM-GX5-25
  • Motor Encoders (Wheel Odometry): Yes
• Description: This bag file is compressed with a command zstd. You can decompress with the command zstd -d utsukuba22_university_of_tsukuba.bag.zst
• License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)

UTsukuba 2022, University of Tsukuba, Map Data

• Short Name: map_utsukuba22_university_of_tsukuba
• Provider (Team): University of Tsukuba
• Location: Tsukuba campus 2022
• Type: Map
• File: map_utsukuba22_university_of_tsukuba.pcd
• Size: 224 MB
• Format: pcd
• Number of Points: 14,656,120
• Point Type:
  • XYZ: Yes
  • Intensity: Yes
  • Color: No
  • Normal: Yes
• SLAM Method: LIO-SAM
• Description: Tsukuba Campus of University of Tsukuba.
• License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)

Tsudanuma 2020, Chiba Institute of Technology, Sensor Data

• Short Name: tsudanuma20_cit
• Provider (Team): Chiba Institute of Technology
• Type: Sensor
• Location: Tsudanuma 2020
• File: tsudanuma20_cit_compressed.bag
• Size: 57 GB
• Format: rosbag
• Date: 2020-08-27 17:43:12
• Duration: 56:12s
• Setup: Mobile Robot (Joystick Operation)
• Sensors:
  • Lidar: Velodyne VLP-16
  • Camera: Intel RealSense D435i (without depth)
  • Radar: No
  • GNSS: Drogger DG-PRO1RW (Independent Positioning)
  • IMU: Analog Devices ADIS1646

搜集汇总

数据集介绍

构建方式

tc-datasets 数据集由多个研究团队在 Tsukuba Challenge 等自动驾驶导航挑战赛中构建，涵盖了不同年份和地点的传感器数据与地图数据。数据集的构建主要依赖于多种传感器，如激光雷达（LiDAR）、惯性测量单元（IMU）、全球导航卫星系统（GNSS）等，结合同步定位与地图构建（SLAM）技术生成高精度地图。部分数据通过差分轮式机器人或移动机器人进行采集，确保了数据的多样性和真实性。数据格式以 PCD 和 ROSBAG 为主，便于后续处理与分析。

特点

tc-datasets 数据集的特点在于其丰富的多源数据融合，涵盖了不同年份、地点和传感器的组合。数据集不仅包含高精度的点云地图，还提供了详细的传感器数据，如激光雷达、IMU 和轮式编码器等。数据集的多样性使其适用于多种自动驾驶导航任务，如路径规划、环境感知和定位。此外，数据集中的地图数据经过移动物体去除处理，确保了地图的静态性和可靠性。所有数据均以开放许可发布，便于学术研究和工业应用。

使用方法

tc-datasets 数据集的使用方法较为灵活，用户可根据研究需求选择特定的传感器数据或地图数据进行下载。数据集以 PCD 和 ROSBAG 格式提供，用户可通过 ROS（机器人操作系统）或点云处理工具（如 PCL）进行数据解析与处理。对于地图数据，用户可直接加载点云文件进行环境建模或路径规划；对于传感器数据，可通过 ROSBAG 文件回放传感器数据流，模拟真实场景下的导航任务。使用该数据集时，建议引用相关论文以支持学术研究的透明性和可重复性。

背景与挑战

背景概述

tc-datasets是由多个研究团队在筑波挑战赛（Tsukuba Challenge）中收集的自动驾驶导航数据集，涵盖了2018年至2023年间的多个比赛场次。该数据集由筑波大学、fuRo团队、MapIV等机构提供，主要包含传感器数据和地图数据，旨在为自动驾驶系统的开发与测试提供真实世界的环境数据。数据集的核心研究问题集中在如何通过多传感器融合和SLAM技术实现高精度的环境感知与定位。这些数据不仅为学术界提供了宝贵的研究资源，还推动了自动驾驶技术在复杂环境中的应用。

当前挑战

tc-datasets在解决自动驾驶导航问题时面临多重挑战。首先，传感器数据的多样性和复杂性要求高效的融合算法，以应对不同传感器（如LiDAR、IMU、GNSS）之间的数据同步与校准问题。其次，地图数据的构建需要克服动态环境中移动物体的干扰，确保生成的地图具有高精度和鲁棒性。此外，数据集的构建过程中还面临数据量大、存储与传输成本高的问题，尤其是在处理高分辨率点云数据时，如何有效压缩和存储数据成为一大技术难点。这些挑战不仅考验了数据集的构建技术，也为相关领域的研究提供了重要的研究方向。

常用场景

经典使用场景

tc-datasets 数据集广泛应用于自动驾驶导航领域的研究与开发。该数据集通过提供高精度的激光雷达（LiDAR）和传感器数据，支持研究人员在复杂环境中进行实时定位与地图构建（SLAM）算法的验证与优化。特别是在城市和校园等多样化场景中，该数据集为自动驾驶系统的路径规划和环境感知提供了丰富的实验数据。

解决学术问题

tc-datasets 解决了自动驾驶领域中的多个关键学术问题，例如如何在动态环境中实现高精度的定位与地图构建，以及如何通过多传感器融合提升导航系统的鲁棒性。该数据集为研究人员提供了真实世界的多源数据，支持对SLAM算法、路径规划技术和传感器融合方法的深入研究，推动了自动驾驶技术的理论突破与工程实践。

衍生相关工作

基于 tc-datasets，许多经典研究工作得以展开。例如，研究人员利用该数据集开发了高效的SLAM算法，提出了基于多传感器融合的动态环境感知方法，并探索了在复杂地形中的路径规划技术。这些工作不仅推动了自动驾驶领域的学术进展，还为相关技术的商业化应用奠定了基础。