A periodic motion pattern generation by recalling well-suited CPG parameters based on time-series observations

This paper proposes a periodic motion pattern generation model inspired by biological brain motor systems. The model consists of three parts, "Brain", "CPG" and "Body." The Brain part is a time-series pattern discriminator modeled by RNN, which works as a predictive selector of CPG parameters. The CPG part is a rhythmic pattern generator for a lower level motor control, is represented by Matsuoka\´s neural oscillator model, and the Body corresponds to the dynamics of physical interactions between controlled systems and environments. In the proposed model, the Brain part can recognize several kinds of environmental changes through its proprioceptive feedback time-series stem from own action, and also it can modulate its motion pattern by recalling well-suited CPG parameters with respect to the current Body dynamics.