| Summary: | This thesis presents the design of a system for robot localization named ROLL. The
ROLL system was designed to be used in a dynamic real-time robot testbed. The system could
provide location information to experimental robot control architectures.
There are many existing localization systems for robots using many different techniques
and sensors. All of these localization systems either require custom computer hardware or are
much too slow or require the robot to stop for a certain period of time for the localization system
to produce output. The system in this thesis allows localization without having to have the robot
stop, and without the need for expensive custom hardware.
The ROLL system uses a camera mounted on the robot facing forward to detect special
landmarks placed in the environment. From the shape of the landmarks in the camera image, the
position of the camera relative to the robot can be calculated. This position can be transformed
into absolute frame coordinates using the known absolute frame position of the landmark.
To test the ROLL system, it was implemented in the ACE programming environment on
a stand-alone workstation. This implementation was fed sequences of stored images that
included a visible landmark to simulate a real robot trajectory. The system was also implemented
on a Texas Instruments TMS320C40 DSP board to prove that the algorithm could be run at
acceptable real time speeds.
The system was able to detect the landmark in all of the images in two series of 11
images each, except for one. The position of the camera was calculated to within 4.2cm of the
known position for each image. The TMS320C40 implementation was able to provide location
information at an average of approximately 8Hz.
|
|---|
