Automatic Tuning of a Mechanical Design Parameter of a Robotic Leg by Iterative Learning Mechatronics

The best control performance in mechatronic systems cannot be obtained by only control algorithms, because the control is effective only within the performance range realizable by an actuator system in practice. To obtain the best control performance, therefore, the mechanical system should be designed considering the control performance first. It is, however, difficult to expect the control performance in the mechanical design process, since the control performance is dependent on various unexpectable factors, such as input and output saturations of an actuator, heat problems, sensor limitations and so on, as well as the mechanical design parameters. Therefore, in this paper a recursive mechanical design process based on control experiments, called Iterative Learning Mechatronics (ILM), is proposed. The proposed method seeks a better mechanical design parameter based on a set of control signals, such as a control input and a tracking error, and suggests an update of the design parameter. For verification of the proposed method, the ILM is applied to obtain the best leg stiffness of a robotic leg bearing a load.