Adaptive neural-network-based control for motion planning of joint-constrained redundant robots subject to unknown disturbances

Journal article, 2026

Redundant robots are widely used in industrial and medical scenarios, where the motion planning of redundant robots is a complex challenge due to kinematic redundancy and various internal and external disturbances. This paper proposes a novel quadratic programming (QP) framework that integrates system modeling, adaptive control, and neural optimization techniques to robustly get the redundancy resolution under several objectives/constraints during the path-tracking tasks. To realize real-time motion planning, an adaptive zeroing neural network (AZNN) is designed, which features exponential convergence and strong noise immunity through a specially constructed activation function. Furthermore, this paper studies the complex uncertainties/noises in the kinematic model of redundant robots. External disturbances are specifically explored and categorized into two types: those generated by unknown exosystems, which are handled via the internal model method, and those generated by known exosystems, which are mitigated by utilizing the Luenberger observer method. These strategies ensure that the AZNN solver maintains optimality and stability even under periodic noise and model mismatch. Unlike existing noise-immune zeroing neural networks (ZNNs), the newly designed AZNN suppresses noise by using the internal model method without introducing the sign(·) function, which may cause chattering problems, guaranteeing robots’ safety. The convergence analysis of the AZNN controller is rigorously carried out, and an example is given to illustrate the effectiveness of our proposed methods.