Unlike rigid parts, flexible parts can be deformed by contact force during assembly. For successful assembly, information about their deformation as well as possible misalignments between the holes and their respective mating parts is essential. However, because of the nonlinear and complex relationship between parts deformation and reaction forces, it is difficult to acquire all required information from the reaction forces alone. Therefore, the effective method that can measure both of parts deformation and misalignments is required. For robot assembly, the corrective motion to compensate for the misalignment has to be determined from the measured information. In this case, however, it can't be simply derived from the measured misalignment alone because the parts deformation continues during misalignment compensation. The relationship between them is very complex.
In this paper, we propose a visual sensing system for measuring parts deformation in any direction and misalignments in flexible parts assembly. This system is composed of a camera and a series of mirrors, and can overcome the self-occlusion problem in which a hole is occluded by its mating part.
Second, we present an algorithm that can measure parts deformation and misalignments in cylindrical peg-in-hole tasks. Simulation and experimental results show that the proposed system and algorithm are effective in measuring parts deformation and misalignments, thereby dramatically increasing the rate of success in assembly operations.
Finally, we analyze the process of flexible parts assembly, and present a neural net-based inference system that can infer the complex relationship between the corrective motion and the measured information of parts deformation and misalignments. Experimental results show that the proposed neural net-based assembly algorithm is effective in compensating for the lateral error, and that the proposed algorithm can be extended to the assembly tasks under more general conditions.