Long-Term Motion Generation Using Deep Neural Network in Imitation Learning

Abstract

Recently, imitation learning has attracted attention as the method for robots to execute tasks in response to environmental changes. Previously, imitation learning using bilateral control showed successful results for tasks requiring force adjustment. However, these studies had difficulty in executing the tasks requiring long-term inference and their verification is not sufficient. Robots must perform long-term tasks for more general tasks. In this study, the performance of the long-term task with a deep neural network (DNN) model was verified. The deep neural network model was expected to extract more long-term features. The validation was performed on the task of writing the letters “ABC”, and the robot could execute the task using a DNN model.