A GIM-based approach for biomimetic robot motion learning

This paper presents a novel GIM-based learning approach for biomimetic robot motion learning. A general internal model is developed for describing discrete and rhythmic animal movements. Based on analysis, the paper shows that the learning approach is able to generate similar movement patterns for the robots directly, through the minimum changes in GIM parameters, thus avoid the usual time-consuming learning or training process. The GIM also exhibits the phase-shift property, which is necessary when the coordination among multiple GIMs is required to perform a complex task. Finally, the GIM-based learning approach is applied to learn two basic fish-like swimming patterns for a biomimatic robotic fish under locomotion control. The results verify the effectiveness of the learning approach.