TeamKAIST DRC Dataset
(last update : 09-MAR-2016)

Welcome! This page is made for sharing the TeamKAIST DRC dataset (download) .  The dataset has been captured for three days (Jun 4~6, including rehearsal) during the DARPA Robotics Challenge Finals 2015. It consists of 130 pairs of images(*.jpg) and 3D point clouds(*.csv). we describes some explanation about our sensor system, 3D point cloud files, and viewer on this page. You can download the dataset and its viewer at the last of this page. If you want to see more details about DARPA Robotics Challenges and TeamKAIST, please visit official DRC homepage and TeamKAIST project page.


Hubo Head System 

     
     DRC-HUBO+ has one camera and one 2D LIDAR sensor installed on its head to obtain color and depth information, respectively. We use a GigE camera by PointGrey and a laser range finder sensor by Hokuyo. As shown in figure on the right side, the camera and the LIDAR are attached on a vertically rotatable rig together so that the LIDAR scans multiple horizontal scan-lines to fill the vertical field of view (FOV) of the camera. 
     The LIDAR horizontally scans 180 degrees with the angular resolution of 0.25 degree. The vertical distance between the scan-lines, which determines the density of the 3D data, can be controlled by the motor speed of the rig.  The robot usually scans about 150k points from -30 degree (top) to +60 degree (bottom) when the horizontal plane is at zero degree. The maximum reliable distance of the LIDAR is 10 meters.


3D Point Cloud Data 


     The file format of our 3D point cloud data is csv(comma separated value).
 It contains the number of 3D points is on the very first line. From the second line, location information of each point is written on the individual line.
     The coordinate of 3D positions(X, Y, Z) is motor coordinate(see above image), minus value(-1) for pixel location(U, V) means that the point exists out of the camera field of view, and LIDAR index(I, J) of a 3D point starts from zero. The end of vertical index(I) depends on the number of scan lines of the scene, and the end of horizontal index(J) is 720.
     The file name of 3D point cloud is the same as the corresponding image(eg. Day0_0001.jpg, if the other is Day0_0001.csv) which grabbed just before the scanning of 3D point cloud when the robot was not moved.
     If you want to check our 3D data, you can use our DRC data viewer program.

Dataset Viewer (Download the viewer only*)
* At least one pair of image(*.jpg) and LIDAR(*.csv) files is requested in the same directory to run this program.
     The viewer is programmed using Microsoft Visual Studio 2013 with OpenCV 2.4.9, OpenGL, and MFC library. The file list which is in the same folder as the program files is automatically loaded when you execute the viewer.

TeamKAIST DRC Dataset Download

 Rehearsal Day
  June 4, 2015 at Fairplex in Pomona, California, USA
43 pairs of image and point cloud
 Day 1
  June 5, 2015 at Fairplex in Pomona, California, USA
54 pairs of image and point cloud
 Day 2
  June 6, 2015 at Fairplex in Pomona, California, USA
33 pairs of image and point cloud


Comments