Neural networks for offline segmentation of teleoperation tasks

The authors present two artificial neural network architectures that perform offline segmentation of telerobotics tasks using force data. Two network architectures have been tested. The first one is based on turning temporal sequences into spatial patterns. The second architecture extends the first model by including the network's output in the input array. Experimental data are classified offline by a hidden Markov model providing the transition times and the corresponding segmentation for the training data. It was found that the first architecture needs a high number of iterations for learning the associations, whereas the latter has a high convergence speed. >