A novel computational approach for robotic spatial perception has been proposed and tested in this study. This model is based on the techniques of nonmetric multidimensional scaling (MDS). From the nonmetric data of rank order of distance between points in a configuration, we can reconstruct the precise coordinates of this configuration from the MDS model.; Various types of configurations, both structured and unstructured, have been simulated. The results proved that the MDS is an accurate and robust model for relative recovery of metric coordinates. The accuracy of the results is amazingly immune to certain types of common measurement resolution limit inherent in either real machine or human "fuzzy" sensors. This means that the measuring device employed for this model need not be precise.; A series of weighted MDS models have been developed to successfully solve the rotational problem of reconstruction. The accurate and robust absolute recovering can be achieved from two proposed models: a classical MDS model with reference points and a retinal model, which reconstructs the configuration from two conjugate recovered images.; A deduction algorithm was developed to enhance the input data. By implementing this data enhancement, a great deal of pair comparisons can be saved for reconstructions with certain accuracy range. It is surprising that, for optimal efficiency, the measuring device cannot be too precise.; The models developed in this study can also somewhat explain some human visual phenomena. A plausible human vision model based on these models may be developed in future.
展开▼
