   
AUTONOMOUS STAIRCLIMBING
Motivation:
For robots operating in man-made environments, the ability to climb stairs is
essential. Especially for vehicles used for emergency response (e.g., to search
for victims in a building), fast and reliable stair-climbing is necessary for
mission success. Teleoperation, which is currently the standard mode of use for
search-and-rescue robots, is not an appealing option when it comes to climbing
stairs. Usually the robot maneuvers outside the field of view of the
operators, forcing them to rely only on visual feedback from the robot. However,
images from the robot's camera are often blurred due to the robot's motion, have
limited field of view, and are subject to latency, due to the communication
channel. These factors, combined with the robot's slippage on the stair edges,
can result in inaccurate and slow stair climbing, collisions with the stair
walls, and even in toppling of the vehicle.
.
Contribution:
In our work, we have
designed an algorithm for autonomous stair climbing with a tracked vehicle. Our
approach relies on fast and accurate estimation of the robot's 3D attitude and
of its position relative to the stair boundaries. An Extended Kalman Filter is
used for attitude estimation, fusing rotational velocity measurements from a
3-axial gyroscope, and measurements of the stair edges acquired with an onboard
camera. A two-tiered controller, comprising a
centering- and a heading-control module, utilizes the estimates to guide the
robot fast, safely, and accurately upstairs. A block diagram of the algorithm is
shown below:
The algorithm has
been implemented on iRobot Packbot (shown above), and has demonstrated robust
performance under real-world conditions. The key factors contributing to the
algorithm's reliability are the high rate at which attitude estimates are
produced (100 Hz in our implementation), and the fact that no assumptions about
the stair geometry, the dynamics of the vehicle's interaction with the stair
surface, or the lighting conditions are imposed. As a result, the system is
capable of robust operation in challenging situations.
Owing to its generality,
the developed stair-climbing algorithm can be implemented on any tracked vehicle
that is equipped with a 3-axis gyro and a camera, and can be employed on robots
used in search-and-rescue missions and military operations, to increase the
mobility of handicapped people, or to improve the efficiency of household
helping robots.
Results:
One of the most challenging tests of the stair-climbing algorithm
took place at the Tampa Police and Fire Training Academy, in Tampa FL. During
the Spring 2005 NSF Industry/University Cooperative Research Center (I/U CRC) on
Safety, Security, and Rescue Research (SSRRC) Spring 2005 Symposium, the robot
was required to autonomously climb the outdoor metal staircase of a fire tower.
This flight of stairs was very steep and slippery, due to wear of the steps'
color coating. Despite these challenging conditions, the robot was able
to overcome the extreme slippage, and to safely reach the top of the stairs. A
video of the robot's camera view during the ascent can be downloaded here:
zipped .avi file (10MB)
The detected straight lines are superimposed
on the video images. Note the severe slippage and jerky
motion that the robot undergoes in the initial phase of the climb. In spite of
the very fast dynamics of the motion, the controller is able to securely guide
the robot up the stairs, due to the high bandwidth of the attitude estimates.
Related Publication:
A.I. Mourikis,
N. Trawny, S.I. Roumeliotis, D.M. Helmick, L.H. Matthies: ``Autonomous
Stair Climbing for Tracked Vehicles,'' Vision and Robotics,
Joint issue of the International Journal of Computer Vision
and the International Journal of Robotics Research
(to appear).
pdf
Acknowledgements:
This work was supported by the University
of Minnesota (DTC), the Jet Propulsion Laboratory (Grant No. 1251073, 1260245,
1263201), and the National Science Foundation (ITR-0324864, MRI-0420836).
Joel Hesch, Le Vong Lo, Faraz Mirzaei, Kyle Smith, and Thor Andreas Tangen
offered their invaluable support during hardware/software development and
experimental testing and validation.
|
