Tracking control of humanoid manipulator using sliding mode with neural network and disturbance observer

Abstract

The nursing robot, equipped with a 6-degree-of-freedom (6-DOF) humanoid manipulator, has been applied in elderly and disabled care to execute complex and random nursing tasks with its advantages in automation and intelligence. Especially, when the nursing robot performs daily care tasks such as serving tea and pouring water, the good trajectory tracking performance of its manipulator is a crucial capability. However, nonlinear coupling, model uncertainty, joint friction, unknown external disturbances, and particularly the fact that manipulator does not satisfy Pieper criterion-are the main challenges, which degrade control performance. Few existing studies have simultaneously addressed all these issues to improve the control accuracy of the manipulator. Therefore, to achieve the good tracking performance for manipulator, a robust control method combining sliding mode control (SMC), radial basis function neural network (RBFNN), and nonlinear disturbance observer (NDO) is proposed. An improved Jacobian-based gradient descent method solves inverse kinematics, with the improved gradient descent driven inverse kinematics (IGDIK) module ensuring accuracy; RBFNN compensates for model uncertainty; NDO handles disturbances and friction. Simulations and experiments demonstrate enhanced trajectory tracking accuracy and stability, validating its effectiveness for the target manipulator.