Tactile recognition based on repetition of grasping of an object with a robot hand (Calculation method for increasing probability levels using fuzzy relations)

摘要

A new robotic tactile recognition method wherein repetitions of grasping of an object enhanced recognition certainty was presented. To evaluate the certainty, probability levels of recognition were calculated by a fuzzy relational matrix defining similarities between objects. Calculations using various kinds of fuzzy compositions and binary operations required for the iterative evaluation process were compared. Arithmetic sum and product were found to be most suitable, being able to reflect the increase in probability levels. By applying the arithmetic calculations, a new program module for increasing probability levels was implemented. The module was incorporated into the fuzzy production system which was composed of a calculation module for characteristic parameters and an inference module using fuzzy production rules. Three modules were cyclically processed through repetitions of grasping of an object. A series of tactile recognition experiments using a robot hand equipped with three-axis tactile sensors were performed. Consequently, the present system succeeded in recognizing three-dimensional shapes of unknown objects such as an icosahedron and an imperfect sphere, which could not be recognized through only a single grasping.