A Hierarchical Control and Learning Network for Redundant Manipulators With Unknown Physical Parameters

This article proposes a hierarchical control and learning (HCL) network to implement multitask hierarchical control and physical parameter estimation for the redundant manipulator. This network is comprised of a learning subnetwork and a control subnetwork. The learning subnetwork is designed for the online estimation of physical parameters based on the state information of the manipulator. Meanwhile, it has an event-triggered mechanism to activate or stop the estimation process. Besides, the control subnetwork can handle multiple tasks and decouple them using null-space projection matrices. This decoupling relationship allows higher priority tasks to be allocated to degrees of freedom (DOFs) first, while lower priority tasks do not interfere with higher priority tasks. Finally, the HCL network is tested on a Franka Emika Panda (FEP) manipulator. Simulative and experimental results demonstrate that our proposed method outperforms existing methods for handling unknown physical parameters and executing multiple tasks.