The Málaga 2009 Robotic Dataset Collection

1. Description

This dataset was collected at the parking of the Computer Science Faculty of the University of Málaga (Spain) using an electric car equipped with 3 SICK and 2 Hokuyo laser scanners, 2 Firewire color cameras, one xSens IMU, three RTK GPS receivers and one consumer-grade USB GPS receiver. As described in the paper, a centimeter-level ground truth is robustly computed from the RTK data, thus making the dataset an ideal testbed for SLAM or localization techniques.

This page contains a description of the dataset, including a detailed specification of the format of all its files. This is only one out of the collection of datasets available for download at the dataset collection page .

Overview of the 3D point-cloud

(click to enlarge image)

2. File format specifications

If you are reading this README file from a downloaded dataset, the following files and directories should be also included:

	malaga_datasets_grabber.ini
	It contains the 6D poses of all the sensors relative to the coordinate system of the vehicle, among other parameters such as the cameras framerate, distortion parameters, etc. The format of this file is that used by the application rawlog-grabber, used to record the dataset.

	<name>.rawlog
	This is the binary log file with all the grabbed data. The format of this file is based on the open-source Mobile Robot Programming Toolkit (MRPT) and it is described in detail here. The rest of the files below are ultimately derived from this log file.

Images

All the images grabbed by the right (CAMERA1_<timestamp>.jpg) and left (CAMERA2_<timestamp>.jpg) cameras during the whole dataset duration. These are the original images captured by the cameras, before rectification. The estimated camera parameters can be found in the paper, in RawlogViewer by clicking on any image or in malaga_datasets_grabber.ini. The image files can be used independently, but their names are set such as the RawlogViewer application or MRPT-based programs can directly access them through the binary rawlog file.

	Images_rect
	All the images grabbed by the right (CAMERA1_<timestamp>.jpg) and left (CAMERA2_<timestamp>.jpg) cameras during the whole dataset duration, rectified to compensate the camera distortion.

GT_path_vehicle.txt

The vehicle path ground truth at 1Hz as given by the RTK GPS devices. The format of this file is plain text, each row containing the 6D pose at a given instant and its uncertainty (its covariance matrix). Each data column contains: TS x y z α β γ C11 C12 C13 C14 C15 C16 C22 C23 C24 C25 C26 C33 C34 C35 C36 C44 C45 C46 C55 C56 C66 TS : Timestamp, a float number, with the same meaning that standard C time_t but with fractional part up to microseconds. x : X coordinate (meters). y : Y coordinate (meters). z : Z coordinate (meters). α : yaw angle (radians). β : pitch angle (radians). γ : roll angle (radians). Cij : Components of the covariance matrix:

GT_path_vehicle_interp.txt

The vehicle path ground truth, interpolated at 100Hz The format of this file is plain text, each row containing the 6D pose at a given instant. Each data column contains: TS x y z α β γ TS : Timestamp, a float number, with the same meaning that standard C time_t but with fractional part up to microseconds. x : X coordinate (meters). y : Y coordinate (meters). z : Z coordinate (meters). α : yaw angle (radians). β : pitch angle (radians). γ : roll angle (radians).

	GT_path_vehicle.bin.gz
	The vehicle path ground truth at 1Hz, in binary format. This file is a GZ-compressed dump of a CPose3DInterpolator object, which can be loaded into a user C++ program using MRPT through the following commands: using namespace mrpt::poses; using namespace mrpt::utils; CPose3DInterpolator path; CFileGZInputStream("GT_path_vehicle.bin.gz") >> path;

	GT_path_<sensor_name>.txt, GT_path_<sensor_name>_interp.txt
	These files have exactly the same format that GT_path_vehicle.txt and GT_path_vehicle_interp.txt, but they describe the path ground truth for different sensors, i.e. the vehicle poses plus the sensor location on the vehicle using the appropriate 6D pose composition.

	GT_path_vehicle_confidence.txt
	The Mahalanobis distance of the measured inter-GPS square distances to their expected mean and covariance. Lower quality values mean a better measure. The file is plain text, with each row being: TS : Timestamp, a float number, with the same meaning that standard C time_t but with fractional part up to microseconds. q : The quality measure.

	W_star_cov_matrix.txt
	The W* matrix: the reference 6x6 covariance matrix of the estimated vehicle path. The variables are [x y z yaw pitch roll], in meters and radians, respectively. This matrix has been used to obtain the uncertainties of GT_path_vehicle.txt.

	pointcloud.txt, pointcloud.3Dscene and pointcloud_img.png
	A 3D point cloud created from the two vertical SICK laser scanners and their ground-truth path. The files are in plain text format (each line has one triplet x y z for a point, in meters), as a 3D scene compatible with the GUI program SceneViewer3D.

3. Full 6D poses

All the rotations in the files above are given as yaw, pitch and roll, according to the following convention:

Next are the estimated positions of each sensor on the vehicle local coordinate frame, optimized as explained in the article:

4. Software

Most of the dataset is provided in very simple formats (image files, plain text files) to enable its management from virtually any user software (MATLAB, etc.). Nevertheless, the open-source applications RawlogViewer and SceneViewer3D allow a greater potential at inspecting and manipulating the dataset.

They can be downloaded as source code, Windows installer or directly from some GNU/Linux repositories (Ubuntu, Debian, Fedora, etc.) as part of the MRPT project.

5. How to cite?

Please, use the following reference if you want to cite this dataset in your works:

Bibtex:
@ARTICLE{blanco2009cor,
    author = {Blanco, Jos{\'e}-Luis and Moreno, Francisco-Angel and Gonz{\'a}lez, Javier},
     pages = {327--351},
     month = {November},
     title = {A Collection of Outdoor Robotic Datasets with centimeter-accuracy Ground Truth},
   journal = {Autonomous Robots},
    volume = {27},
    number = {4},
      year = {2009},
      issn = {0929-5593},
       url = {http://www.mrpt.org/Paper:Malaga_Dataset_2009},
       doi = {10.1007/s10514-009-9138-7}
}