To realize intelligent robot, it is necessary to establish an effective method for visual and tactile recognition. Much of the existing work on tactile recognition has assumed that tactile sensors derive only data on object surfaces so as to generalize recognition algorithm. With tactile sensors which provide local features of contact areas, e. g, vertex, edge, etc., it is possible to prune inconsistent interpretations more effectively than existing methods.
This paper proposes a method for object identification using local feature sensors. In the method, interpretations at each step of identification process are represented as a set of subsets of 3-D space where a sensor to be located. With the progress of identification process, the set will be narrower. Consequently local constraints become strict and inconsistent interpretations will be efficiently pruned. In the pruning, local constraints are compared considering the error bounds of measurement in sensor data.