Most animal locomotor patterns present cyclic movement, which are produced by the Central Pattern Generator (CPG) which is situated in the spinal cord. Although the CPG is capable of generating periodic motor commands for usual locomotion, it has been reported that not only the CPG but also a higher center such as the cerebellum responds to a perturbation for the locomotion, which suggests that they control locomotor patterns cooperatively. Here we have two questions. One is how two motor commands from the CPG and a higher center work without conflict. The other question is how two control systems, the CPG and a higher center, acquire the motor commands to realize target motions. To solve these questions we propose a hierarchical learning model of the CPG and a higher center. This model assumes that a higher center learns an appropriate motor command which works as a supervisory signal for the CPG to learn an appropriate firing pattern. We applied this model to the learning control of a one-dimensional hopping robot. As a result, the CPG and a higher center learned desired control signals to realize a target motion, and after learning the CPG was able to generate an appropriate signal for the target motion by itself without receiving signals from the higher center. These results well accord with physiological data observed in many animals.
展开▼
