For the intelligent robot systems, the uncertainty handling capability is a crucial element to accomplish a given task successfully in uncertain situations. Although various sensory information has been used for many intelligent robot systems, most of sensory observations inevitably possess inherent uncertainties caused by the measurement errors, limitations of the operating ranges of sensors, and dynamic situations of the robot environment. For more reliable and robust operations of a robot system in uncertain situations, the synergistic use of multisensory information(multisensor fusion) is needed to get the more trustworthy information about its surroundings.
Over the years, various approaches for the sensory data fusion have been proposed with the purpose of getting more trustworthy information. However, the traditional approaches, which are mainly originated in probability theory, have many drawbacks and limitations such as: no reasoning process under uncertain situations, no consideration of the vaguely defined relative importance between sensory data, not to take account of the effects of the uncertainty of the robot coordinate frame itself, rigorousness that stems from the theoretical basis of the probability theory, etc.
In this thesis, we develop a fuzzy oriented methodology to get some more trustworthy information about the attributes of the robot environment using a fuzzy weighted average and fuzzy reasoning. We describe any geometric primitive of the robot environment as a parameter vector in parameter space. Not only ill-known values of the sensor measurement data and parameterized geometric primitives but the uncertain quantities of coordinate transformations are represented by means of fuzzy numbers restricted to appropriate membership functions. Also we describe the spatial relations between geometric primitives using a simple graph. To get the global information about the robot environment, the correspondence problem between local information is solved using a fuzzy similarity measure and a graph matching technique. Corresponding sensory data combination is carried out using a simple fuzzy arithmetic by taking the subjectively defined degree of relative importance between sensory data into consideration. Also the synergistic use of sensors which have different modalities and characteristics is drawn via fuzzy reasoning using the knowledge and experiences obtained from some experimental study about sensors. As an illustrative example, an experiment is performed on a moving sensor system using a CCD camera and ultrasonic sensor for the recognition of an unknown indoor environment of a robot system.
This methodology is supposed to be useful for many robotic application areas especially in: involving many subjective information, having no exact mathematical models of sensors and environment, operating in dynamic situation in which robust operation is required, using different kinds of sensors simultaneously.