Motion Generation Using Bilateral Control-Based Imitation Learning With Autoregressive Learning

Imitation learning has been studied as an efficient and high-performance method to generate robot motion. Specifically, bilateral control-based imitation learning has been proposed as a method of realizing fast motion. However, the learning approach of this method leads to the accumulation of prediction errors during the prediction process and may not generate desirable long-term behavior. Therefore, in this paper, we propose a method of autoregressive learning for bilateral control-based imitation learning to reduce the accumulation of prediction errors. A new neural network model for implementing autoregressive learning is also proposed. Three types of experiments are conducted to verify the effectiveness of the proposed method, where the method is shown to have improved performance over those of conventional approaches. Due to the structure and method of autoregressive learning employed by the developed model, the proposed method can generate desirable long-term motion for successful tasks and has a high generalization ability for environmental changes based on the human demonstrations of tasks.