Human error identification in programming by demonstration of compliant motion robotic tasks

In this paper, we suggest a scheme for error identification in human skills transfer when using the Programming by Demonstration (PbD) in adding a set of skills from a human operator to the force- controlled robotic tasks. Such errors in human skills transfer is majorly caused from the difficulty of properly synchronizing the human and machine responses. Based on the captured Cartesian forces and torques signals of the manipulated object, we present an approach of identifying the errors stemmed from human wrong skills transfer in a PbD process. The scheme is composed of using the Gravitational Search- Fuzzy Clustering Algorithm (GS-FSA) in finding the centroid of the captured forces and torques signals for each Contact Formation (CF). Then using a distance- based outlier identification approach along with the centroid of each signal, the human errors can be identified in the framework of data outlier identification. In order to validate the approach, a test stand, composed of a KUKA Light Weight Robot manipulating a rigid cube object, is built. The manipulated object is assumed to interact with an environment composed of three orthogonal planes. Error identification for two case studies will be considered and other cases can be dealt with in a similar manner. From the experimental results, excellent human error identification is shown when using the suggested approach.