Novel Noise-Tolerant Recurrent Neural Network Adaptive Control of Preoperative Pose for Cranial Neurosurgery Robot With Physical Constraints

The preoperative pose manual adjustment of the cranial neurosurgery robot (CNR) has defects such as low operation efficiency and great difficulty, and the research on the adaptive pose adjustment technology of CNR can effectively compensate for these defects. However, the existing neural network control technology of robots often neglects issues such as noise-tolerant, computational burden (CB), control accuracy, and physical constraints, which are crucial to the stable work of CNR. To this end, a novel noise-tolerant recurrent neural network (RNN) control method with physical constraints is designed and successfully controls the CNR to achieve self-adaptive adjustment of the preoperative pose. First, based on the CNR’s kinematics model, a quadratic programming control scheme with physical constraint is designed, and then a new noise-tolerant RNN is designed to resolve the CNR’s control scheme. Under noise-free and noise conditions, theoretical analyses are rigorously performed to ensure the fast convergence of the designed neural methods. Finally, the practicability and correctness of the designed neural methods are verified by using a general numerical test case and the CNR. The experimental analysis results show that the designed neural control methods can well control the CNR to complete the self-adaptive adjustment of the preoperative pose and has the ability of antinoise interference. Compared with the prior art, the designed neural methods have lower CB, stronger noise-tolerant ability, and higher convergence accuracy.