DOC PREVIEW
huber_daniel_2003_1

This preview shows page 1-2 out of 6 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 6 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 6 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 6 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

The 4th International Conference on Field and Service Robotics, July 14–16, 2003Automatic 3D underground mine mappingDaniel F. Huber Nicolas VandapelThe Robotics InstituteCarnegie Mellon UniversityPittsburgh, Pennsylvania [email protected] [email protected] several years, our research group has been de-veloping methods for automated modeling of 3D environ-ments. In September, 2002, we were given the opportunityto demonstrate our mapping capability in an undergroundcoal mine. The opportunity arose as a result of the Que-creek mine accident, in which an inaccurate map causedminers to breach an abandoned, water-filled mine, trap-ping them for several days. Our field test illustrates the fea-sibility and potential of high resolution three-dimensional(3D) mapping of an underground coal mine using a cart-mounted 3D laser scanner. This paper presents our experi-mental setup, the automatic 3D modeling method used, andthe results of the field test. In addition, we address issuesrelated to laser sensing in a coal mine environment.1 IntroductionFor several years, our research group has been devel-oping methods for automated modeling of 3D environ-ments [3][4][5]. In September, 2002, we were given theopportunity to demonstrate our mapping capability in anunderground coal mine, the Mine Safety and Health Ad-ministration (MSHA) research mine in Bruceton, Pennsyl-vania. The opportunity arose as a result of the Quecreekmine accident in July, 2002, in which miners inadvertentlybreached an abandoned, water-filled mine, trapping them-selves amidst thousands of tons of water. After the min-ers were safely rescued, an investigation was launched todetermine the cause of the accident and to identify newprocedures necessary to prevent mine breaches in the fu-ture. Regulations already in place aim to prevent such anaccident: mapping the mine before ending operations, ex-ploratory drilling, and so forth. Unfortunately, old mapsmay be incorrect, incomplete, or simply lost. In theend, the Quecreek accident was attributed to an inaccuratemap [2].A collaborative effort by several research groups atCarnegie Mellon University (CMU) has been formed to de-velop robots to autonomously map abandoned mines andactive mines before operations are ended. Such robotswould be an important contribution to mining safety. De-tails can be found in [1, 14, 9]. In this paper, we addressthe problem of sensing and generating high-resolution 3Dmodels of an active mine. In September, 2002, we con-ducted a field test to demonstrate the feasibility of highresolution 3D mapping of an underground coal mine us-ing a cart-mounted 3D laser scanner. The remainder of thepaper is organized as follows. First, section 2 reviews pre-vious work on mine mapping and localization. Section 3describes our experimental setup and the data collectionprocess. Section 4 explains our automatic modeling algo-rithm and discusses the resulting 3D model. Finally, sec-tion 5 presents an analysis of the issues relevant to lasersensing in a coal mine environment.2 Related workIn this section, we review the most relevant work onmine mapping and localization. Early work by Shaffer [13]described a method to localize a mobile robot in an un-derground mine by registering terrain features (corner andline segments) extracted from an a priori survey map withcross-sections from an environment map produced by alaser scanner. In [11, 12], Scheding extensively tested a setof navigation sensors mounted on a Load, Haul, and Dumptruck (LHD) in the harsh underground mine environment.Using the data from a laser line scanner coupled with thenavigation data of the vehicle, he produced a 3D modelof a section of the mine. In [7], two line scanners wereintegrated on an LHD. The iterative closest point (ICP) al-gorithm was used to register the 2D profiles to an existingmap. This implementation was extended to mine mappingin [8]. The contributions presented above focused on vehi-cle automation for active mines. In the context of mappingFigure 1: The cart-mounted Z+F laser scanner used in thedata collection.abandoned mines, Thrun [14] produced 2D maps and par-tial 3D models of tunnels, using a SLAM approach withtwo line scanning lasers mounted on a tele-operated robot[1]. Several systems have been designed to map mines thatare inaccessible to a ground robot, for example, by map-ping a cavity using a 3D laser sensor inserted through abore-hole. Such systems include the C-ALS (Cavity Au-toscanning laser system) by Measurement Devices, Ltd.and the Cavity monitoring system by Optech, Inc.1A sim-ilar approach has been followed in [9].3 Data collectionFor our field test, we used a high resolution 3D laserscanner mounted on a cart as illustrated in figure 1. Thesensor, a Zoller and Fröhlich LARA 25200 (Z+F) scanner[6], produces 8000×1400 pixel range and reflectance im-ages with millimeter-level accuracy. The field of view is360◦×70◦with a range of 22.5 m. The laser scan headwas inclined to allow higher density scanning of the floorand ceiling near the scanning platform. Unfortunately, insome regions, the low roof was actually too close to thescan head for the sensor to fully scan the ceiling.We obtained 23 scans at three- to five-meter intervalsalong a loop trajectory through a sequence of 4 hallways(figure 2). The cart was kept stationary at each location forthe 90 seconds required to obtain each scan. Due to the ca-pabilities of our modeling algorithms, it was not necessaryto record the position or attitude of the cart. This greatlysimplifies the data collection process. The entire procedureonly took about three hours, including setup and disassem-bly of the equipment. For this experiment, the cart wasmoved manually, but it would be straightforward to mountthe scanner on an autonomous mobile robot.1www.mdl.co.uk, www.optech.on.ca, May 200350’50’Figure 2: Surveyed map of the Bruceton mine. The redcircle indicates the area mapped in our field experiment.4 Automatic modeling from realityModeling-from-reality is the process of creating digitalthree-dimensional (3D) models of real-world scenes from3D views as obtained, for example, from range sensors orstereo camera systems. Recently, we have developed a sys-tem that fully automates the modeling-from-reality process[4][5]. The key challenge of automatic modeling-from-reality is the accurate and robust registration of multiple 3Dviews. Although each input scan is an accurate representa-tion of the 3D structure of


huber_daniel_2003_1

Download huber_daniel_2003_1
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view huber_daniel_2003_1 and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view huber_daniel_2003_1 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?