The cerebellum plays a pivotal role in motor control and learning. Due to its uniform neuronal network structure throughout the cerebellar cortex among different animal species, the cerebellum is considered as a general motor learning machine that is capable of controlling wide varieties of control objects by the common neuronal network structure and learning algorithm. We have constructed a physio-anatomically realistic cerebellar neuronal network model, and implemented on a PC so that it can run in real-time and work as an adaptive controller of robots. Here we summarize our cerebellar model from its structure and learning algorithm to soft and hardware implementation, and show examples of application to robot control.

Artificial general intelligence (AGI) is able to acquire knowledge through interaction with the environment. It is expected to have the ability to respond appropriately to situations that were not assumed at the time of design. However in performance evaluation that focuses only on task performance ability, it becomes easier to build knowledge about each task in question, departing from traditional AGI research. Accordingly, to assess AGI based on the knowledge that has been built from the performance evaluation can be expected to have an inhibitory effect on the creation of so called Big Switch Statement type programs. In this study, a basic examination of how well suited a program is to perform a task by looking at its length and its performance in solving problems in a domain. Specifically, sorting programs written in the Java language are examined considering their code lengths and execution time. An evaluation method was constructed and applied to the sorting programs.