Planetary Pits and Caves Analog Dataset Banner

[Description] [Support] [Data Download] [Scripts] [Attribution] [Acknowledgements]


Notice (Mar 18, 2019) - We have moved! This dataset was formerly hosted on Carnegie Mellon University servers as the Planetary Pits and Caves 3D Dataset. A change in data archival policy and relocation of the original author (Dr. Uland Wong) to NASA Ames Research Center resulted in the removal of public access. As the data was funded through NASA grants, it was decided to find a new permanent home on NASA servers. An added benefit is that there are now resources to provide maintenance and updates from our new work. 

The CAVES project at CMU developed robotic capabilities for exploration and 3D mapping of pits and skylights, which are giant holes on the surfaces of planets which may lead to the scientific holy grail of intact lava tubes. Over a two year period, starting in September 2012, many terrestrial analog sites were surveyed with high resolution LIDAR scanning in support of robot testing, algorithmic development and ground truthing. These analog sites either represent the size and shape of skylights/pits on planetary bodies or are geologically significant in their own right. As a part of end-of-project dissemination, this dataset is being released to the public to facilitate robotics and exploration research in this domain.

Data presented here are point cloud models which were created using a high-end LIDAR survey scanner, typically from the FARO line (x130, x330). The scanner is stationary and mounted on a tripod. Many view points and scan locations are stitched to create a unified model of the site. Manually selected features were typically utilized for rough matching followed by ICP for fine registration. A robotic total station, which tracks the scan location, was sometimes utilized to facilitate this process. The models are colorized utilizing the onboard camera of the scanner under natural illumination. As sequential scanning often took hours, the illumination may have changed significantly between parts of the unified model. We have taken steps to normalize brightness across the scans utilizing naive voxel-based histogramming. Other color correction schemes are left to end user implementation. Several sites were also imaged with a DSLR camera for photographic data quality beyond the sensor's onboard camera.

Examples of successful research based on this data are available here, here and here.

Dataset Changelog

Mar 18, 2019 - Relocation of data to NASA and info update

Feb 03, 2015 - Release of Indian Tunnel data

Dec 28, 2014 - Addition of Fieg Highwall volume model

Dec 18, 2014 - Addition of Sheepridge volume model

Dec 15, 2014 - Initial release of dataset

Support and Disclaimer

This data is offered as a courtesy to researchers. No guarantees are made regarding the content, quality or accuracy of any file downloaded from this website, so use at your own risk! For technical questions, please email the current maintainer Uland Wong <uland dot wong at nasa dot gov>.

Data Download

The dataset is partitioned into several downloads, one for each analog site. Data for each site contains *.f32 point cloud files. Refer to the accompanying matlab scripts for examples of how to read the .f32 data files. Data is stored as sequential, 7-attribute readings, in binary little-endian float32 format. There are no delimiting characters or values.

The format of the .f32 file is:
    [x_coord][y_coord][z_coord][NIR_reflect][R_color][G_color][B_color]     #Point 1 attribs
    [x_coord][y_coord][z_coord][NIR_reflect][R_color][G_color][B_color]     #Point 2 attribs
                                                                       ...                                                          #Point n attribs

The first three attributes are the Cartesian coordinates of an individual point in the cloud. Units of the point clouds are in meters. Colors are stored in attribute #5-7, and range from [0,255]. They are either RGB or grayscale (R=G=B). If infrared reflectance data was recorded, it is stored in attribute #4, otherwise zeros are recorded.  

To aid in understanding of this data, we have also included a screen shot of the point cloud and a color photograph of each site. These files are distributed as .rar archives.


King's Bowl - COTM


3D Point cloud of King's Bowl pit

Photograph of King's Bowl Pit


This dataset from Craters of the Moon National Monument and Preserve covers the King's Bowl pit and immediate vicinity. King's Bowl is a phreatic explosion pit caused by groundwater meeting lava, causing a steam explosion approximately 2200 years ago. The pit is located on the park's Great Rift and is 100m long, 30m wide, and 30m deep. This area was mapped in July 2014 by scanning with a Faro X330 from 44 stations located around the pit rim over 3 days. Scans were timed in the morning and afternoon such that natural sunlight illuminated the side of the pit being mapped. This dataset includes two models: a cleaned version and the original output. The cleaned model removes many artifacts and crops to just the pit itself but is downsampled in resolution from the raw data. The original version preserves native resolution and leaves all artifacts (e.g. ghosting people and equipment) intact. It includes scans of the terrain surrounding the pit.

Download Kingsbowl.rar [530 MB]

Indian Tunnel - COTM


3D point cloud of indian tunnel

photograph of indian tunnel

Indian Tunnel is the largest of the lava tubes in the caves area of Craters of the Moon park. It was formed by the Blue Dragon Flow 2100 years ago. The tunnel is over 10m in diameter and 250m long. Portions of Indian tunnel were mapped in August 2014. The tube features numerous collapses and skylights which were mapped from the surface using the same approach as King's Bowl. Critical portions of the interior were also mapped, though it would have been prohibitive to map the entire length. Special illumination was designed for the scanner to enable quality color data in the dark underground portions. A special multi-view dataset featuring ultra-resolution scanning and HDR photography of a single 4m x 4m piece of the wall is included in the download. Due to the many files and the expansive nature of this dataset, we have also posted a supplementary data overview document below.

Download IndianTunnel Surface and Collapses .rar [1.1 GB]

Download IndianTunnel Cave Model .rar [2.07 GB]

Download IndianTunnel HDR Wall Photos .rar [175 MB]

Download COTM_Supplemental_Info.pdf


Fieg Site


3D point cloud of fieg site

photograph of fieg site


The Fieg site was a surface coal mine in Somerset County, PA. These artificially shaped pits were used for early robot testing and mapped several times in late 2012 through early 2013. We have found that the cornered geometry of the Fieg site is similar to at least a few examples of impact melt pits on the moon. Moreover, the intricate layering of geology on the walls is particularly interesting. Due to the ephemeral nature of strip mining, these sites no longer exist. The dataset archive contains several models of the site. The highwall model consists of a rectangular high resolution crop of layered wall geology, meshed and ready for 3D printing. 

Download Fieg.rar [243 MB]

Download FiegHighwallVolume.rar [38 MB]

Sheepridge Site


3d point cloud of sheepridge site

photograph of sheepridge site

Sheepridge was a medium-large surface mine pit In Johnstown, PA area. It was mapped in Sept 2013. As with all surface mines, it is a man-made structure with short-lived features. This particular site was used for final demonstration of our Tyrobot explorer prototype. Features of interest for Sheepridge include great depth and smooth ramps from the rim to the floor from several routes. The ramp features were utilized to study how a wheeled rover might negotiate the ramp of the Lacus Mortis pit on the moon. In addition to the standard point cloud, a watertight mesh model that can be 3D printed is available for download.

Download Sheepridge.rar [328 MB]

Download SheepridgeVolume.rar [11 MB]


We have included several Matlab files to aid in use and understanding of this dataset. Two utility files for reading and writing *.f32 files are write_pts_binary_float32.m and read_pts_binary_float32.m. The format is otherwise very simple and intended for painless I/O in any language. We encourage conversion of point data to more efficient proprietary formats once downloaded and extracted. Some simple functions for downsampling point clouds and operating on voxels have also been included to get started.

Download Matlab scripts [7 KB]


This dataset can only be used for research or academic purposes. Any commercial use of the data whatsoever or incorporation of the data into a larger database intended for public distribution must be done with the explicit written consent of the dataset authors. Until then, please use the following attribution if you use the data:

Uland Wong, Warren Whittaker, Heather Jones, Red Whittaker. NASA Planetary Pits and Caves Analog Dataset. December, 2014.


This work was supported through NASA Innovative Advanced Concepts (NIAC) grant NNX12AQ55G. We are immensely grateful to Faro Technologies, Inc. for loan of an X330 scanner used in data collection.

We acknowledge the work of Warren "Chuck" Whittaker, Heather Jones, Ander Solorzano, Chris Cunningham and Wennie Tabib. We thank PBS Coals Inc., Fieg Bros Coal Co., the National Park Service at Craters of the Moon, and NASA Ames Finesse SSERVI team for access and use of test sites. 


NASA Official: Terry Fong <>
Curator: Intelligent Robotics Group
NASA Privacy Statement