The purpose of this dissertation is to clarify the system configuration method that enables rapid deployment in order to quickly reconstruct the robotic assembly system in response to the introduction of new products. Specifically, we propose three methods as following: (1) an automatic training dataset generation method for a quickly trainable recognition system, (2) a method for generating an assembly sequence using only an assembled CAD model, and (3) an assembly method using a general purpose flexible jig inspired by a jamming gripper.
Using a mechanical product including several rigid parts of various shapes and a deformable part, experiments were conducted to evaluate in terms of the versatility and accuracy of each method and the time required to reconfigure the system. The results prove that the training dataset collection time can be drastically reduced, feasible and easy-to-assembly sequences are generated, and fixing and assembling various parts with the proposed flexible jig is possible.