Pedaling by a redundant musculo-skeletal humanoid robot

We aim to realize a humanoid system which can do natural motions like human. Pedaling a cycle is a very essential motion including many problems for a humanoid system, such as mutual interaction with environments, optimization of load distribution and so on. We insist that a humanoid should have more redundancy, flexibility and complexity to solve these problems. But, the more redundant a humanoid's body becomes, the more important a software system of a humanoid control is. We present a system which can acquire its own body information by a self repeated training of the real humanoid and can be used for realization of pedaling of a human-like redundant and complicated musculo-skeletal humanoid. This paper discusses general problems of pedaling a cycle and the required specifications of the self-training system and describes the hardware setup required for pedaling. We also carried out some experiments to show the feasibility of this system.