Optimizing Redundant Manipulator Performance: A Dual-Criteria Control Approach via Dynamic Neural Networks

Current neural network-based control schemes for redundant manipulators predominantly focus on controlling a single performance metric. For more options in performance control, this paper proposes a dual-criterion control strategy based on an untrained dynamic neural networks algorithm (UDA). This strategy takes into account physical constraints as well as orientation tracking. Additionally, this approach is transformed into a standardized quadratic programming (QP) problem, with a global optimum sought using a UDA solver. Moreover, simulations based on the Franka Emika Panda manipulator in MATLAB and CoppeliaSim validate the feasibility and flexibility of the proposed scheme. Compared with existing schemes, this approach demonstrates superior capabilities in balancing and optimizing multiple performance metrics.

I. Introduction

Redundant manipulators, characterized by extra degrees of freedom (DOFs), are capable of executing complex tasks [1]–[3], thereby gaining widespread application in fields [4] such as industrial production and control. To meet the demands of industrial tasks, redundant manipulators have seen rapid development in sub-tasks such as dynamic obstacle avoidance [5] and tracking control [6], [7].