At the 2014 International Conference on Intelligent
Robots and Systems (IROS) in Chicago, IL, USA, six teams from around the world
took up the challenge of navigating an Adept MobileRobots Pioneer 3-DX mobile
robot in a difficult, novel environment using only a Kinect for Windows RGB-D
camera as a sensor, in combination with the mobile robot’s own motion sensing
and position estimate. This contest was organized and sponsored by the robotics
group from Microsoft Research, and Adept MobileRobots (http://www.mobilerobots.com).
Nine qualifying teams were chosen for
the contest, and began working on their navigation implementations 13 weeks
before the contest. In this short time, the teams had only general information
about the environment in which the robot would have to navigate.
Six teams were present at IROS: “Maxed
Out” (SV-ROS, USA), “SCARAB” (Seoul National University, Korea), ONERA
(France), MetroRobotics (USA), “Kordis” (Korea Advanced Institute of Science
and Technology, Korea), “Blackberry” (Virginia State University, USA).
Contestants’ software had to use the
Kinect to sense the objects in front of the robot, as visual images and/or distance
measurements; communicate with the Pioneer 3-DX robot to drive it and receive
measurements of its movement as sensed
by its internal motion controller and wheel encoder sensors; create a
representation or understanding of the environment including specific goal
locations within it; and communicate with the external benchmark system used to
determine whether the robot had reached the desired goal locations.
This contest provided many challenges in
robot navigation and software engineering, and pressed the participants skills
in autonomous robot mapping, navigation and localization as well as some practical
constraints seen in real applications of robotics. Contestants noted that the Kinect presents
limitations in field of view and stability of data when compared to other, more
expensive sensors such as scanning laser (LIDAR) sensors. They also noted the time pressure to perform
tests and reconfiguration and debugging when finally on-site just days and
hours before the start of the contest. Teams had to be agile in adjusting their
software for specific details of the environment such as the appearance of the
carpet, walls and furniture. Some features in particular were especially challenging,
including very narrow furniture, and a long narrow hallway that provided few
unique localization features. An
important aspect of the challenge was the presence of people during navigation
trials that were not present during the earlier mapping and preparation stages. Most teams worked late into the night before
the contest.
All teams used existing ROS (http://www.ros.org) as a
foundation for their systems. ROS is a
large open-source project for robotics software, and provides a system of
reusable components that provide standardized interfaces to hardware such as
the Kinect sensor and the Pioneer 3 DX robot, components for performing
mapping, localization, and planning and navigation of the robot through a
mapped 2D space, as well as visualization and other development tools. Teams customized and added to this foundation
of existing code to implement their unique approaches to the challenge.
The first place winning team, “Maxed Out”,
representing the Silicon Valley ROS users group, used ROS to incorporate Mathiew Labbe’s
Rtabmap mapping and localization algorithm, which used both the Kinect’s
distance measurement capability and its visual camera to sense obstacles as
well as match features previously recorded in the map during the mapping and
preparation phase of the contest. Team
members present at the contest collaborated with each other and remote members
of the team to test, debug and improve their total system, and their work
resulted in the highest score with 3 waypoints successfully visited in the
fastest amount of time of 80.4 seconds.
More details on this entry are available from the team at http://www.meetup.com/SV-ROS-users/pages/Winning_the_IROS2014_Microsoft_Connect_Challenge/ and https://code.google.com/p/rtabmap/wiki/IROS2014KinectChallenge.
Second place was claimed by Charles
Lesire for the French National Aerospace Research Laboratory (ONERA), also
successfully reaching 3 waypoints, in a time of 157.6 seconds.
The third place finisher, David Lu of
Metro Robotics, using his modified ROS navigation stack, reached 2 waypoints in
a time of 33.9 seconds, followed by Kordis (2 waypoints, 82 secs.), SCARAB (2
waypoints, 97 secs.) and Blackberry (1 goal, 32 secs).
The members of the winning
team split a grand prize of $5,000 USD.
All runner up competitors also received a prize of a new
second-generation Kinect One for Windows Developer Kit.
(Photo Credit to George Maxwell)
