Abstract
© The Author(s), under exclusive licence to Shiraz University 2025.This paper addresses a fundamental question in motion planning for slender manipulators: how can an optimal tip-point trajectory be determined while accounting for collision risk across all points of the manipulator's body hull without proximity sensors? Additionally, is it feasible to develop an offline-trained point-to-point motion strategy that minimizes the risk of collision? An enhanced surrogate model based on a matrix of multi-layer perceptron neural networks, designed to estimate the nonlinear vibration zones surrounding the desired configuration of the manipulator hull is the proposed solution. The system is structured around fundamental components, including a single-type link, revolute joint, and cable, whose mechanical simplicity and slender form inherently lead to high-amplitude fluctuations that must be mitigated in point-to-point movements. A key limitation of conventional optimization approaches lies in their inability to achieve real-time responsiveness due to the complex interaction between obstacle detection, multi-degree-of-freedom system dynamics, and control mechanisms. This study introduces a novel surrogate model trained for rapid and reliable response, enabling real-time path optimization based on kinematics and geometry while integrating neural networks to estimate vibration zones associated with specific kinematic configurations. The surrogate model, trained offline to incorporate dynamic effects, ensures comprehensive risk assessment across the manipulator hull for all point-to-point tasks. By leveraging reinforcement Q-learning, the proposed approach facilitates efficient real-time motion planning, delivering a fast and viable solution for collision-free path optimization.
NETL
NSTL
Springer Nature