http://www2.science.uva.nl/sites/cogniron/dataSet10000plus/readme.txt This directory contains a dataset acquired by a mobile robot driving through an indoor environment. The robot was equipped with an omnidirectional vision system, a laser range scanner, sonar range scanners and odometry sensors. Because of the large loop in the building (+/250 meters) the dataset is an interesting and challenging test case for SLAM-algorithms based on Sonar, Laser, Omnidirectional Vision, or a combination of these sensors. --------------------------------------------------------- The dataset is split into three zip-files: filename filesize contains images.zip 622 MB set10000/images/image00?????.jpg set10000/imtimes.txt set10000/imcalib/image000??.jpg laser.zip 33 MB set10000/laser.txt odosonar.zip 1.4 MB set10000/odometry.txt set10000/sonar.txt A brief description of the files contained in the zip-files (see below for more details): images/image000???.jpg all omnidirectional images imtimes.txt an ASCII file containing the timestamps of the images imcalib/image000??.jpg omnidirectional images of a calibration pattern for calibration purposes laser.txt raw laser range data and timestamps odometry.txt odometry data and timestamps sonar.txt raw sonar range data and timestamps --------------------------------------------------------- Detailed description: The dataset was taken by driving a Nomad Scout II robot around in an office building of the University of Amsterdam and the Nikhef institute. The robot was tele-operated using a remote usb joystick. It drove through long corridors occasionally entering neighboring rooms, as if it was autonomously exploring the environment. The dataset is challenging for a couple reasons. -The building consists of a large loop of +/- 250 meters. The robot's trajectory includes this loop, coming back in the same part of the environment after 45 minutes of driving. -The environment was not completely static. Sometimes persons walk by the robot and doors open and close as one would expect in an office environment. -The environment, and especially the corridors, are quite dark, darker than typical office environments. The light intensity sometimes measured only 30 lux, as measured by a Sekonic light meter (which is comparable to a living room lit by Christmas tree lights). Because the shutter time of the camera needs to be kept low to prevent motion blur, the acquired images have relatively low contrast and the high gain results in some salt and pepper noise. -The environment also includes some large open spaces (e.g. entrance halls). Sometimes, if the robot drives in the middle of such open spaces, the range scanners do not reach any of the surrounding walls, making the range data useless. --------------------------------------------------------- Image-data The omnidirectional vision system consists of a Dragonfly2 camera from Point Grey Research (http://www.ptgrey.com) which is a 1 mega pixel (1024X768) firewire camera and a convex hyperbolic mirror from Accowle (http://www.accowle.com/english/), positioned in such a way that the resulting vision system has a single view point (see http://staff.science.uva.nl/~obooij/publications/Zivkovic05OmniCam/index.html for a more detailed explanation of this vision system). The focal length of the lens mounted on the camera was chosen such that the mirror occupies a large part of the view of the camera. Due to the rectangular shape of the CCD, some small parts of the mirror, namely at the right and at the left of the robot, are not seen in the image. Accowle, the company that manufactured the mirror, provided us with the parameters that describe the shape of the mirror which is needed for the reconstruction of image rays given an image pixel: a = 42.0882 b = 25.0915 elevation = 0.4887 radians depression = 0.9599 radians rim diameter = 62 mm distance F-F'= 116 mm The center of the mirror in the camera image and the focal length of the camera can be obtained by various calibration tools available on the web such as Christopher Mei's "Omnidirectional Calibration Toolbox" (http://www.robots.ox.ac.uk/~cmei/Toolbox.html). Calibration images are available in imcalib.zip. For our experiments we used the following parameters: x-pos = 530 pixels y-pos = 389 pixels x-focal length = 1739.62724 y-focal length = 1741.40106 For a detailed description of these parameters and how to use them for ray reconstruction have a look at http://staff.science.uva.nl/~obooij/publications/Zivkovic05OmniCam/index.html. The camera axis of the omnidirectional vision system is approximately aligned with the robot center of rotation. The center of projection of the omnidirectional vision system is positioned at +/- 114 cm from the ground plane. In total the imageset consists of 10325 gray-scale images. The raw images use more than 8 GB. To save band-width the images are provided in jpeg-format, produced using ImageMagick with a jpeg quality of 80%. Still, the resulting zip file images.zip with the jpeg images is 620 MB. If you have trouble downloading this file or you have interest in the raw images send an email in order to arrange a different file transfer. During each image acquisition a timestamp was saved in an associated file called imtime.txt. The columns of these files hold the following information: column description 1 timestamp of time of measurement, milliseconds since 1970 2 name of the image file --------------------------------------------------------- Laser range data At the front of the robot a Hokuyo URG-04LX Laser Sensor was mounted facing in the driving direction and scanning horizontally. This laser range scanner is particularly useful for indoor environments because of its small size and weight and its wide scan angle (+/- 240 degrees). One of the drawbacks is its relatively small sensing range of +/- 4 meters. The raw range measurements as well as the timestamps for each measurement is located in laser.txt in the following manner: column description 1 timestamp of time of measurement, milliseconds since 1970 2 laser range scans in meters at -276.48/2 + 0*(0.36) degrees 3 laser range scans in meters at -276.48/2 + 1*(0.36) degrees 4 laser range scans in meters at -276.48/2 + 2*(0.36) degrees ... ... 385 laser range scans in meters at -276.48/2 + 383*(0.36) degrees 386 laser range scans in meters at 0 degrees 387 laser range scans in meters at -276.48/2 + 385*(0.36) degrees ... ... 768 laser range scans in meters at -276.48/2 + 766*(0.36) degrees 769 laser range scans in meters at -276.48/2 + 767*(0.36) degrees 770 laser range scans in meters at -276.48/2 + 768*(0.36) degrees The laser sensor was positioned +/- 65.5 cm from the ground, +/- 11 cm to the front of the center of rotation of the robot, and +/- 0 cm to the right of the center of rotation of the robot. Column 2 holds the extreme right scan, column 386 holds the scan pointing in the driving direction of the robot (to the front), and column 770 holds the extreme left scan. Not all scan data is valid. This is in part because of the low intensity of the laser which causes invalid readings when the reflection is poor. Also, most of the scan data at the extreme angles are invalid. We found that the scan values should be interpreted as follows: scan value interpretation 0 no reflection detected (maximum range of 4 meters exceeded) <0.019 invalid scan (but very similar to scan value 0) 0.019 invalid scan (found at the extreme angles) all others valid scan in meters --------------------------------------------------------- Odometry data On average 12 odometry measurements per second were taken. Because the robot has solid wheels the odometry is quite accurate. The columns of odometry.txt hold the following information: column description 1 timestamp of time of measurement, milliseconds since 1970 2 odometry, x coordinate in meters 3 odometry, y coordinate in meters 4 odometry, orientation in radians 5 odometry, translational velocity in meters per second 6 odometry, rotational velocity in radians per second The coordinate system used for the robots odometry is the standard Cartesian coordinate system where 0 orientation corresponds to the direction of the x axis. The orientation increases in anti-clockwise direction. All odometry information is relative to an arbitrary fixed point. --------------------------------------------------------- Sonar data The robot is equipped with 16 ultrasonic sonar sensors which give a 360 degrees range scan. The sonar sensors do not measure simultaneously. Each sonar sensor starts measuring after the previous sensor has finished processing which in turn depends on the time it takes for the sound to come back with a maximum of 200 milliseconds. The sonar sensors are read out simultaneously with the odometry sensors at a frequency of 12 Hz. Therefore, the sonar measurements include a lot of data that was actually measured in previous time steps. All sonar range information is relative to the orientation of the robot where 0 radians corresponds to the forward direction of the robot. Radians increase in the anti-clockwise direction. All sonar cells are mounted on a ring with radius 0.19 meters centered over the center of rotation at a height of 0.26 meters. The columns of sonar.txt hold the following information: column description 1 timestamp of time of measurement, milliseconds since 1970 2 sonar range information at 0/16 * 2Pi radians in meters 3 sonar range information at 1/16 * 2Pi radians in meters 4 sonar range information at 2/16 * 2Pi radians in meters 5 sonar range information at 3/16 * 2Pi radians in meters 6 sonar range information at 4/16 * 2Pi radians in meters 7 sonar range information at 5/16 * 2Pi radians in meters 8 sonar range information at 6/16 * 2Pi radians in meters 9 sonar range information at 7/16 * 2Pi radians in meters 10 sonar range information at 8/16 * 2Pi radians in meters 11 sonar range information at 9/16 * 2Pi radians in meters 12 sonar range information at 10/16 * 2Pi radians in meters 13 sonar range information at 11/16 * 2Pi radians in meters 14 sonar range information at 12/16 * 2Pi radians in meters 15 sonar range information at 13/16 * 2Pi radians in meters 16 sonar range information at 14/16 * 2Pi radians in meters 17 sonar range information at 15/16 * 2Pi radians in meters --------------------------------------------------------- This dataset was taken with the help of Bas Terwijn, Isaac Esteban and Olaf Booij. It was conducted within the EU Integrated Project COGNIRON ("The Cognitive Companion"), funded by the European Commission Division FP6-IST Future and Emerging Technologies under Contract FP6-002020. We would appreciate it to receive an email if this dataset has been useful to you. Also, for any questions regarding the dataset we can be reached by mail: bterwijnATscienceDOTuvaDOTnl obooijATscienceDOTuvaDOTnl iestebanATscienceDOTuvaDOTnl Intelligent Systems Lab Amsterdam Informatics Institute Faculty of Science University of Amsterdam Kruislaan 403 1098 SJ Amsterdam The Netherlands