Structural Design of Force-Sensing Limbs and Quasi-Static Interaction Force Estimation for a 6-RUS Parallel Manipulator

Aiming at the drawbacks of the end-mounted six-axis force/torque sensor scheme for parallel manipulators in environmental interaction, including vulnerability to machining dust erosion and occupation of end-effector workspace, as well as the limitations of the end-sensor scheme for human-robot compliant operation with narrow interaction range and the high-cost characteristic of current-based sensorless force estimation, this paper takes the 6-RUS parallel mechanism as the research object and proposes an end-effector-free force-measurement scheme with sensors embedded in the US links. Single-axis tension/compression load cells are coaxially embedded within six US links. Bearing fit and tolerance constraints are adopted to suppress lateral load disturbances, so that the tension/compression load cells only acquire the axial forces of the links. The quasi-static static mapping relationship of the mechanism is established based on screw theory. Taking static nonlinear disturbances such as link self-weight, joint clearance and assembly errors into consideration, a multilayer perceptron (MLP) is constructed, whose inputs are six revolute joint angles and six-channel sensor readings of limb tension/compression load cells, and whose output is the six-dimensional interaction force at the end-effector. Thereby, the interaction force estimation at the end-effector of the quasi-static 6-RUS parallel mechanism is realized. The experimental results demonstrate that when the moving platform of the parallel mechanism keeps static or moves at low constant speed, the proposed scheme combining force-sensing limbs and the MLP can realize real-time estimation of external interaction forces exerted on the end-effector with low training cost. It provides a low-cost and high-robustness solution of limb-embedded sensors for parallel manipulators oriented to environmental interaction and human-robot interaction.