Redundant manipulators, endowed with multiple degrees of freedom, are capable of executing several tasks simultaneously. However, the mutual interaction among these tasks can severely degrade overall performance. To address this issue, a priority-based recurrent neural network (PBRNN) is proposed for hierarchical control of redundant manipulators. This method combines the traditional RNN with a null-space projection matrix to decouple high-priority tasks from low-priority ones. When the total required dimension of the task exceeds the manipulator's degrees of freedom, low-priority tasks are omitted. In addition, a priority-switching recurrent neural network (PSRNN) is devised based on the PBRNN to address control discontinuity caused by task priority switching. By introducing a priority-switching compensation term, it effectively mitigates excessive output signals during priority transitions. Relevant convergence analyses, along with simulations and physical experiments, validate the feasibility and superiority of the proposed methods for handling multi-task execution and priority-switching scenarios.
Xie et al. (Sun,) studied this question.