Sensorless End-Effector Force Estimation of Multi-DOF Manipulators Using Semiparametric Joint Model and Generalized Momentum Observer

With end-effector force estimation, manipulators can safely interact with the environment or operators, eliminating the need for conventional wrist sensors that measure force and torque. This paper introduces a new approach to estimating the force exerted by the end-effector, utilizing a two-step method. The first step is to build an accurate semiparametric joint model. The joint model integrates a harmonic drive parametric model with a local Gaussian process regression (LGPR)-based learning model. A combined parametric model of the harmonic drive offers the analytical joint torque component, which is employed in the LGPR process to reconstruct the unmodeled portion. This reconstruction is based on the motor-side joint angle, the link-side joint angle, and the controlled motor current. Compared to the parametric or learning models alone, this semiparametric joint model considerably improves modeling precision. In the second step, an external force observer is developed by estimating joint torque. The force observer developed in this paper is based on generalized momentum. Multi-DOF manipulators benefit from this design because joint dynamics are separated from link dynamics, thereby reducing the complexity of dynamic modeling. In addition to joint torque sensors, a wrist force/torque sensor is utilized to compare the estimated joint torques and end-effector forces with the actual measured values. Experimental results obtained using a custom-built manipulator have demonstrated the advantages and effectiveness of this approach.