Multiresolutional Filter Application for Spatial Information Fusion in Robot Navigation

Abstract

Robot navigation is one of the major fields of study in autonomous robotics (Oriolio, G., Ulivi, G. et al., 1998; Beetz, M., Arbuckle, T. et al., 2001; Wang, M. & Liu, J.N.K., 2004). In the present work, application of a multiresolutional filter for spatial information fusion in vision robot navigation is described. The novelty of the research is the enhanced estimation of the spatial sensory information in autonomous robotics by means of wavelet decomposition into multiresolutional levels and the fusion of the processed information while the processing takes place at the respective levels. Although wavelet-based information fusion is used in different applications (Hong, L., 1994; Hsin, H.C. & Li, A.C., 2006), its application in robotics is not common in literature. A wavelet-based filtering application in perceptual robotics is articulated earlier where human perception was central to the research. In the present work, optimal vision signal estimation for an autonomous robot is presented, where vision information is obtained by appropriate shaping of the visual sensory information and optimal estimation is obtained by a multiresolutional filter. One of the peculiarities of the research is essentially the wavelet-based dynamic filtering rather than static wavelet decomposition for image analysis or off-line decomposition for signal analysis. A multiresolutional dynamic filtering is a rather complex system. In this system vector form of wavelet decomposition is required as briefly mentioned by Hong (Hong, L., 1993) where the description of a multi-sensor fusion rather than wavelet decomposition was central to the study. In contrast to (Hong, L., 1993), in this work the multiresolutional dynamic filtering is central to the study together with the emphasis on application peculiarities in autonomous robotics; namely, several options for vector-wise operations are pointed out and one of the options is described in detail. Also, in autonomous robotics, the estimation of angular velocity is not a measurable quantity and it has to be estimated from the measurable state variables so that obstacle avoidance problem is taken care of. Therefore, the angular velocity estimation in real-time is a critical task in autonomous robotics and from this viewpoint the multiresolutional spatial information fusion process is desirable for enhanced robot performance. For the fusion process extended Kalman filtering can be used and is actually used in the present study. Perceptual vision employed in this study is one of the vision alternatives in robotics providing several essential possibilities characterizing the robot as to its performance. Especially perceptual robotics is gaining importance in the last decade as to its expected connection to human-like behaviour. The organization of the work is as follows. O pe n A cc es s D at ab as e w w w .ite ch on lin e. co m