Success/Failure Identification of Skill Movement by Neural Network Using Force Information

Currently, in the field of FA(Factory Automation), automation has been accomplished for only tasks with high reproducibility. Meanwhile, there remain a range of areas where tasks cannot reliably be performed due to their complexity or larger environmental variation. In these cases, the reliability of the task can be improved by recognizing the task failure through success/failure identifications and executing the task again when it fails. However, the success/failure identification using conventional machine learning methods has not been discussed for determining the success or failure for unlearned objects. Thus, this paper examined assembly tasks and demonstrated that the success or failure for an unlearned object can be identified by taking advantage of generalized nature of the neural network using force information. The results of making success/failure identifications using information on force, image, and position were compared and the advantage of force information in tasks was demonstrated.

[1]  Huosheng Hu,et al.  Robot Learning from Demonstration in Robotic Assembly: A Survey , 2018, Robotics.

[2]  Kensuke Harada,et al.  Cantilever snap assemblies failure detection using SVMs and the RCBHT , 2014, 2014 IEEE International Conference on Mechatronics and Automation.

[3]  Yoshua Bengio,et al.  Gradient-based learning applied to document recognition , 1998, Proc. IEEE.

[4]  Jürgen Schmidhuber,et al.  Long Short-Term Memory , 1997, Neural Computation.

[5]  Marc'Aurelio Ranzato,et al.  Building high-level features using large scale unsupervised learning , 2011, 2013 IEEE International Conference on Acoustics, Speech and Signal Processing.

[6]  Wyatt S. Newman,et al.  Interpretation of force and moment signals for compliant peg-in-hole assembly , 2001, Proceedings 2001 ICRA. IEEE International Conference on Robotics and Automation (Cat. No.01CH37164).