1239
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
as not navigable 12.6% of the time with a height threshold
parameter of 5=15 cm [6].
I
Figure 5: An illustration of fused sensor maps
performance was maximized. Terrain data can be from maps or
aerial imagery, but high fidelity and accuracy are achieved by
using POS data to register laser range scans into models that we
call “drive-by topography.” These models are obtained by
driving a vehicle equipped with laser scanner and POS system
over terrain and recording topographic imagery. The method is
broadly applicable for detailed surveys that are unachievable
from satellite or aerial flyover [7].
Detailed terrain topography can be acquired by collecting range
scanner and vehicle position measurements while driving. This
was done with an HI Hummer called ‘Topographer’ which
utilized a POS LV and laser scanner to derive drive by
topography typically with ,25m resolution and 1.5m accuracy.
This data is combined to generate a height map reconstructed by
solving for the position of each range measurement in 3-D space.
The resulting surface models provide resolution and accuracy
that are unobtainable from satellites or from traditional maps.
An example of the detail of topography is shown in Figure 7.
Planning and Vehicle Control
With reliable data from the POS LV integrated into the drive-
by-wire systems of both Red Team Robots, pure pursuit
tracking was made possible. However a method to maximize
the performance of both vehicles was needed.
Figure 7: Topography data overlaid on imagery
Human drivers adjust to changing terrain / weather conditions in
addition to interpreting a curves apex to maximize the
efficiency of a turn rather than following a straight line denoting
the curve and ‘jerking’ through it. This is not efficient and
providing the robot with apex entry and exit information, in
addition to terrain condition, are two ways in which
Figure 8: Result of pre-planning process
The entire robot preplanning process relies on accurate terrain
and known parameters of vehicle performance to detail safe
driving parameters while minimizing the time it takes to
complete a section of the course [8]. The result of pre planning
is illustrated in Figure 8. The black lines denote raw RDDF file
waypoints and speed limits provided by DARPA. The red
dotted path illustrates the route as edited by human planners
heavily interpolating the original set of waypoints. These
smoothed splines form the basis of navigating in and out of
curves. Should obstacles be encountered, the robot generates its
PA Coarse Waypoints:
path segment
Processed
Path
Spline Curve
Segment
Figure 6: Illustration of how small errors in position and orientation can provide erroneous terrain characterization