Neural admittance control based on motion intention estimation and force feedforward compensation for human–robot collaboration

Abstract

To enhance robotic manipulator adaptation to human partners and minimize human energy consumption in human–robot collaboration, this paper introduces a neural admittance control framework, which integrates human motion intention estimation and force feedforward compensation. Maximum likelihood estimation is employed to derive human motion intention and stiffness within human–robot collaboration, which are seamlessly merged into admittance control. Force feedforward compensation is proposed to minimize interaction force and position tracking fluctuations based on estimated human intention and stiffness. RBF neural network control is used to refine position inner loop dynamics and to improve position tracking accuracy and response speed. Comprehensive comparative simulations validate the method’s effectiveness in diminishing human–robot interaction force, enhancing position response speed, and mitigating interaction force and position fluctuations. The experiment performed on the Franka Emika Panda robot platform, illustrates that the proposed method is effective and enhance human-robot collaboration.