|
Home
US&R
Competitions and Arenas
Reference
Datasets
Rules
Qualification
Process
2007
Events
2006
Events
2005
Events
2004
Events
2003
Events
2002
Events
2001
Events
2000
Events
FAQs
Publications
Contact
|
|
Special
versions of NIST's Reference Test Arenas for Urban Search and Rescue
Robots have been developed to provide the research community with
an efficient way to test their algorithms without having to incur
the costs associated with maintaining functional robots and traveling
to one of the permanent arena sites for validation and practice. These
special arenas consist of real sensor datasets and simulated environments.
Sensor Datasets
The sensor datasets allow programmers anywhere to access and use a
variety of actual sensor data readings gathered inside the NIST arenas.
A systematic grid under the yellow arena allows precise placement
of sensors over grid nodes to represent robot sensor views facing
north, east, south, and west inside the arena. These sensor readings
are compiled into a database allowing easy extraction for perception,
planning, or other sensor-based algorithms. So far one sensor has
been used - a line-scan LADAR, but other advanced (even developmental)
sensors are being considered such as the so-called "flash LADAR".
In this way, a single sensor can produce situational datasets for
anybody interested in algorithm testing, even before the sensors are
widely available or cost effective. The resulting sensor-based algorithms,
when shown to be effective in navigating the virtual datasets, should
have a high likelihood of success when the actual sensor is deployed
in the real arenas. This may occur during competitions or other testing
opportunities in a permanent arena.
Two separate
datasets namely discrete and continuous are posted below.
The discrete datasets contain data corresponding to fixed sensor data
locations whereas in the continuous case, data was collected as the
robot traversed the environment. The full ground truth for
a particular arena configuration is released several weeks after the
data are made available.
Discrete
Dataset in the Yellow Arena
Figure
1:The yellow arena maze for the discrete dataset consists of non-flat
flooring and maze walls. Future datasets in this arena will include
intermittent pitch/roll ramps and open face/top box stacks containing
simulated victims, and continuous pitch and roll ramps.
Continuous
Datasets from RoboCup German Open 2007
Robot Sensor and Command Stream Recordings
Access to data
that was collected in the competition arena of the Rescue Robot League
during the RoboCup German Open at Hannover Fair in April 2007 is available
from http://robotics.iu-bremen.de/datasets/RoboCupGermanOpen2007.
It consists of recorded streams of raw sensor data and motor commands
of a so-called Jacobs rugbot (rugged robot). The data
and the related software can be freely used for academic purposes.
Please use references to the Jacobs robots when you generate own work
based on this data. Please contact Robotics at Jacobs University (http://robotics.jacobs-university.de)
if you want to use the data or the related software for non-academic
purposes. We thank the German Research Foundation / Deutsche Forschungsgemeinschaft
(DFG) for their support to the project 3D Mapping in Unstructured
Environments during which parts of the recording and play back software
for this data were developed.

Continuing
and Future Work
Yellow Arena
All
robots will begin their mission in the Yellow arena maze to directly
compare navigation, victim identification, and mapping capabilities
in autonomous robots along with remote situational awareness and
mapping capabilities in remotely teleoperated robots. The yellow
arena flooring will include 10-degree pitch/roll ramps filling the
hallways between the walls and can either have mostly flat flooring
(covered with paper) with intermittent pitch/roll ramps or mostly
pitch/roll ramps. Within the maze there will be a variety of sensory
obstacles for:
-
ultrasonic
range sensors (absorptive ceiling tiles and reflective corner
angles),
-
laser
range sensors (absorptive dark felt, reflective mirrors, and transparent
plexiglass),
-
victim identification
sensors (false sources of heat, motion, and/or sound).
Orange
Arena
The Orange arena
maze will include continuous 15-degree pitch/roll ramps and half-cubic
(orange) stepfields (see below).
Figure 2: Orange
arena maze with more complex non-flat flooring contains A) continuous
pitch/roll ramps, B, C) half-cubic stepfields and box stacks with
simulated victims that require directed perception and variable illumination
to see inside.
Red
Arena
The Red arena
maze will include continuous full-cubic (red) stepfields as flooring.
There will also be elevated floors accessible by stairs (40 degree
incline, 20cm step heights, 5 steps total) and a steep ramp (35-45
degree incline with carpet for traction) to challenge power and center
of gravity issues. Each elevated floor section will be separated by
a 20cm step/pipe combination to challenge robots reliant on the sharpness
of step edges for traction and reward robots that can change their
shape to reliably surmount curbs of any condition. Confined spaces
under the elevated floors will have stalactites hanging down and may
be placed over ramps or stepfields.

Figure 3: Red
arena maze includes A,B) full-cubic (red) stepfield hills,
flats and diagonals, along with C) elevated
floor sections accessible via stairs or a steep ramp, with stalactites
hanging underneath to form confined spaces.

Figure 4: Other
Red arena elements include A) stairs with 40 degree incline and 20cm
step heights, a 35-45 degree ramp with carpet to access the elevated
floors (not shown), B) 20cm step/pipe combinations to minimize corner
traction divide elevated floor sections, and C) confined spaces under
elevated floors (also used with ramp flooring).
|