The number of robots applied to different jobs in our society has
steadily increased over the past years. While industrial robots welding
and varnishing our cars are quite common these days, the future
generation of robots will enter our everyday life. In contrast to
industrial robots working in special evacuated areas for security
reasons, these mobile service robots will have to move among and
interact with humans, imposing new challenges on the software of the
robots. Only the application of sensors to observe the environment and
the subsequent use of intelligent data processing algorithms will enable
the robots to avoid collisions with moving obstacles, to get information
on their current locations, to perform goal oriented tasks and to
cooperate with humans as well as other robots. This thesis addresses
these new challenges by describing the development of a team of robots
that is able to exhibit intelligent behavior, competitive as well as
cooperative, while moving in a highly dynamic environment, as a first
step towards a new kind of mobile robots. To evaluate and compare the
results of the development, robotic soccer (RoboCup) was chosen as
testbed for the robot team. Many of the challenges imposed by the
RoboCup environment are similar to those found in real-life
applications. Robots in RoboCup have to localize themselves, gather
information on their environment, including the position and also the
velocity of potential obstacles, deal with varying lighting scenarios,
as well as competing and cooperating with other robots. Firstly, the
thesis presents the decisions made concerning the co-design of the robot
hardware and software. Here, the selection of an omni-directional drive
mechanism and a vision sensor with a high frame rate, the software
framework designed to be flexible and expandable, as well as the
development of a common and simple control interface are the key points
in the system design. The main part of the thesis, however, presents the
efficient algorithms implemented for image processing, environment
modeling, and high-level control. As the only sensor of the robots is an
omni-directional vision system that is able to map the complete
surrounding of the robot, the performance of the robot system depends on
the image processing algorithms and requires a high accuracy and a low
processing time. The presented image processing algorithms fulfill both
requirements proven by extensive experimental results. Besides landmark
and object extraction these algorithms also include new techniques for
automatic camera calibration. The environment modeling contains a new
combined Monte-Carlo localization and tracking algorithm that is
competitive with the best performing algorithms concerning the accuracy
and the computation time at the same time, again proven by extensive
experimental results. Finally, the high-level robot control component
exhibits efficient movement through highly dynamic environments and
successful competitive and cooperative behavior. The thesis concludes
with an analysis of the team's RoboCup competition results as a final
verification of the good performance of the developed system.
