Contact State Recognition for Selective Cutting Task of Flexible Objects

Robotic machining is typically performed on homogeneous materials. However, achieving autonomous machining of heterogeneous materials could broaden the scope of application of robotic technology. To selectively process only a part of tissues without damaging other tissues, it is necessary to recognize the deformation of heterogeneous objects appropriately. In this study, we developed a method that can recognize materials and deformation types (i.e., elastic/plastic) of flexible objects. Since the features depending on materials and deformation types were observed in time-series response, the proposed identification method is based on a time-delay neural network (TDNN) which exhibits excellent performance in processing and extracting the features of time-series data for discrimination. We verified the usefulness of the TDNN model experimentally by means of (1) identifying flexible objects (i.e., distinguishing between tofu and agar), (2) identifying the elastic and plastic deformations of tofu and agar, and (3) identifying objects, deformation types, and free motion (straight-line) of the robot during the cutting task. The experimental results confirmed that the TDNN model can identify flexible objects and their deformation types with 90% accuracy by learning time-series information of the force and position.