3-D Sensing for Flexible Linear Object Alignment in Robot Cell Production System

Abstract

We have constructed 3-D sensing system for alignment of connector-fitted cables as flexible linear objects which used to be difficult to be automated at the production sites. In the system an industrial robot has a 3-D sensor and a monocular camera mounted at the hand. 3-D sensor, using space encoding method, allows the robot to make high-precision measurements of the order of sub-millimeters, but emphasis is placed on precision at the expense of fields of view. In addition, active sensing methods such as the space encoding method is hard to take measurements for black cables, as well as it has some difficulties with measurements of semitransparent plastic connectors depending on view-points. To cope with those problems, our system is such that the monocular camera on the robot is moved for motion stereo to take measurements on cable shapes; connector’s poses are coarsely estimated from the measurement results; and such view-points as will ensure stable measurements are computed by space encoding method to take precision measurements of connectors. Technical features of the system could be summarized as follows: 1) Determination of view-points to measure connectors, based on measurements of cable shapes, requires no more than two measurements, without repeated searches, to grab semitransparent plastic connectors. 2) Performance of stereo correspondence for plain or black cables, which tends to result in a failure with the aids of no more than irradiated slit patterns and epipolar constraints, has been improved through sequential correspondence inmotion image sequence and its stability evaluations. At the operation tests in the validation system, the robot is assigned a task to assemble the cables into industrial servo amplifiers available on the market, in which automatic alignment of 200 connector-fitted cables has successfully been accomplished in succession to confirm constant performance of the system.