Basic Concepts of the Control and Learning Mechanism of Locomotion by the Central Pattern Generator

摘要

In this chapter, we introduced the basic concepts of the control and learning mechanism to realize a desired locomotion by the CPG. As introduced in section 5.1, the control system of locomotion in living bodies assumes a hierarchical structure in which the CPG synchronizes with a physical system and generates motor signals to motor neurons, and higher centers learn a desired motor signal from the performance of locomotion and control the CPG. The control signal from higher centers can also function as a teacher signal for the CPG to acquire neural parameters in the CPG. The concept that the higher centers acquire a motor command appears to be a natural assumption because higher centers such as the motor cortex sends motor commands in order to respond to sensory stimulus, e.g., avoiding obstacles. Such hierarchical and multiple control system by higher centers and the CPG also contribute to acquire robust control, as shown in the simulation result in section 5.1. The concept of this model with learning rules for coupled oscillators that send signals to each actuator can be applied to a robot with multi-degree of freedom. However, in order to consider the learning system for locomotion, we should consider the learning mechanism not only for the CPG but also for higher centers. For instance, it appears that higher centers play an important role in shaping the motor command that is generated by the CPG, as mentioned in section 5.2. Another important problem is the explosion of the search space of a motor command, e.g., as the number of actuators increases, that is, as the degree of freedom of a robot increases, the search space for the desired motor command becomes vast. Therefore, it is important to consider an efficient learning mechanism for higher centers to find a desired motor command for each actuator from the performance of locomotion. It has been suggested that locomotor patterns of living bodies are well optimized on energy cost; as mentioned in section 5.2; therefore, the minimization of energy cost would be a constraint