Terrain Mapping Using Laser Rangefinders

Perception research in Carnegie Mellon's Planetary Rover program-out of which the walking robot, Ambler, was developed-enables walking robots to create computerized maps of unpredictable, outdoor terrain. Natural terrain is particularly challenging to map, because it does not contain the straight edges and constant lighting of indoor or industrial environments-areas for which perception and mapping technologies have already been developed. To ensure that a planetary rover's perception system will work in unstructured terrain, Carnegie Mellon researchers have tested the rover's perception system over hundreds of different terrains, through changes in lighting level, dust, temperature, and texture of terrain. The perception system was able to navigate the Ambler through these terrains.

Sensing. The rover's perception system views terrain through a laser rangefinder mounted on top of the robot, scanning terrain ahead of the rover within a 60-degree field of view. The rangefinder uses spinning and nodding mirrors to scan the laser beam over the terrain to produce range images. A range image directly measures the distance between the rover and visible points on the terrain ahead. Image preprocessing, which refines the laser image, compensates for "illusions" in lighting and land texture. Darker objects, for example, absorb more laser energy, produce a weaker return signal from the laser, and cause the object to appear farther away. The image preprocessor detects and deletes those pixels that appear unusually far away.

Constructing maps. From the range images, the rover's perception system produces elevation maps. Planetary rovers like Ambler use elevation maps both for locomotion-how and where the rover will place its next step-and for navigation-determining at any moment where the rover is on the landscape, planning a course, and commanding the rover where to go. Elevation maps like the one in Figure 1 allow the rover to plan where and how to place each step without colliding with obstacles in the terrain.

An algorithm in the map-building system shadows regions that are obstructed by intervening terrain. The algorithm also finds and records those regions that lie outside of its field of vision. This way, when the robot's planner asks the perception system for a broader image, the algorithm can report that certain fields of view in the planner's request are not visible. The same algorithm produces elevation maps at whatever resolution the robot's planner asks for.

Compensating for errors. The robot's perception system detects unexpected elements in the terrain produced by the unpredictable effects of lighting, temperature, and texture on the terrain. The robot detects errors by constantly calibrating the difference between its internal terrain map and the height of each foot on the ground. Ultimately trusting what it feels over what it sees, the robot uses a feedback control loop to change its elevation map according to the difference between what it saw and what its legs currently feel from the terrain. By improving the accuracy of the robot's maps, elevation error compensation improves the robot's ability to walk.

Merging maps. The perception system also builds a larger map mosaic out of the smaller map images created from different viewpoints. Generally, the map mosaic algorithm pieces together the smaller maps from the newest to the oldest mapped image. Older map data is used when the navigation planner routes the rover through currently occluded areas that were previously visible. Figure 2 is a map mosaic created from 125 range images acquired at an outdoor site.

While the Ambler walked over kilometers of outdoor terrain over many days of operation, the perception system performed on large amounts of terrain data, successfully producing image after image, map after map.

Point of Contact:
Eric Krotkov
Robotics Institue
Carnegie Mellon University
Pittsburgh, PA 15213 412-268-3058
epk@cs.cmu.edu

Telerobotics Program page

Please email the site webmaster with any comments, criticisms or corrections for this page.
Maintained by: Dave Lavery
Last updated: May 10, 1996