Repetitive Control of Robotic Joint With Variable Impedance Utilizing Agonist-Antagonist Muscle Pair Structure Based on Virtual Trajectories

Realizing accurate trajectory-tracking control for flexible robotic joints using a model-free approach is challenging. In this study, the repetitive control based on virtual trajectory, which is a model-free control method proposed for flexible human multi-joints, is extended and applied to a real robotic joint with a variable mechanical impedance mechanism. The virtual trajectory is modeled by a time series of the equilibrium position of the spring. In conventional methods, the virtual trajectory is corrected according to the trajectory error resulting from repetitive control. To achieve trajectory tracking even with a soft joint, this study proposes a strategy for starting repetitive control with a high joint impedance and switching to a lower joint impedance after convergence to the target trajectory, thereby successively correcting the estimated impedance change ratio by repetitive control. We developed a robotic joint with an agonist-antagonist muscle pair structure that can easily change its mechanical impedance using a position-controlled motor combined with a grommet, which changes its stiffness nonlinearly with respect to elongation, in addition to a wire traction mechanism as the control target. The results of simulations and real robot experiments demonstrated that the proposed method can achieve trajectory-tracking control even when the joint impedance is so low that the trajectory diverges from that of the conventional method. These results show that the proposed method is capable of trajectory tracking control while maintaining joint softness even when the dynamics of the control target and the impedance values before and after the change are unknown, that is, model-free.