Virtual Torque Sensor for Low-Cost RC Servo Motors Based on Dynamic System Identification Utilizing Parametric Constraints.

We propose a novel virtual torque sensor for commercial low-cost radio-controlled (RC) servo motors. The virtual torque sensor has played an important role for conventional robots. It has been used for torque-required control applications such as human⁻robot interaction and under-actuated robots. However, most virtual torque sensors are based on the inversion of actuators or robot dynamics with the assumption that entire dynamics are known. This is not applicable to the RC servo motors that have unknown control structures. As RC servo motors enable researchers and hobbyists to create lightweight but high performance robots in an easy and cost-effective manner, the development of a virtual torque sensor for these motors is necessary. In this study, we propose a design method of a virtual torque sensor for RC servo motors. First, the virtual sensor is derived mathematically based on internal dynamic models with parametric constraints and compared to the conventional model. Second, a dedicated system identification method is developed for the proposed virtual sensor to implement the sensor in actual experiments. Finally, we compare experimental results with the measurements obtained by an actual sensor.

Direct identification of generalized Prandtl-Ishlinskii model inversion for asymmetric hysteresis compensation.

In this study, we present an identification-based direct construction of the inverse generalized Prandtl-Ishlinskii (P-I) model to facilitate inverse model-based feedforward compensation of asymmetric hysteresis nonlinearities. Compared with the derivation of the inverse model analytically from a generalized P-I model, this direct modeling approach has the following advantages. First, direct inverse model identification is formulated as a nonlinear optimization problem, which is not subject to the constraint condition on the generalized P-I model's threshold and density functions, where this is indispensable for the analytical model inversion procedure. Second, this approach may be a simple and attractive alternative when the identification precision of a generalized P-I model is limited by the constraint condition, which necessarily results in insufficient hysteresis compensation functionality for the analytically derived inverse model. Finally, direct inverse model identification can overcome the drawbacks of the analytical inversion method, including the accumulation of parameter estimation errors in an analytical inverse model because these parameters are computed from the generalized P-I model's parameters in a recursive manner. Our experimental results demonstrated that the implementation of open-loop control with the directly identified inverse generalized P-I model as a feedforward compensator achieved precise compensation for the asymmetric hysteresis nonlinearities of a piezoelectric stack actuator.