Computer Engineering Group
Research in the Computer Engineering Group led by Prof. Dr.-Ing. Ralf Möller focuses on two major areas. In the first area, we are developing methods for the visual navigation of mobile robots. Our special approach are insect-inspired “holistic” methods for local visual homing. Holistic methods match entire images instead of local features. They are fast since they can be efficiently parallelized using vector instructions of modern CPUs. We also developed an extensive framework for the navigation of cleaning robots which combines our holistic homing methods with topological or topological-metrical maps. The applicability of this framework was successfully demonstrated in robot experiments in unmodified apartments. We are currently transferring this framework to outdoor environments for the navigation of lawn-mower robots where additional challenges like strong illumination changes and uneven ground have to be addressed. UV-based segmentation of images and fast navigation methods based on spherical harmonics are developed for this purpose. In the second area, we are investigating behavior-based models of visual perception. In computer simulations and robot experiments we attempt to demonstrate that fundamental levels of visual perception – such as an understanding of object shape – could be based on an internal sensorimotor simulation process. Neural network techniques, especially unsupervised methods like local PCA, are developed and applied to train the required internal sensorimotor models (forward and inverse models). We could already demonstrate aspects like saccade control, grasping to extrafoveal targets, and perception of affordances.