控制理论(社会学)
计算机科学
空格(标点符号)
控制工程
控制(管理)
冗余(工程)
控制系统
稳健性(进化)
缩小
鲁棒控制
操纵器(设备)
弹道
执行机构
输出反馈
可控性
数学优化
加速度
机器人
控制器(灌溉)
理论(学习稳定性)
最优控制
自适应控制
车辆动力学
作者
Huanwei Zhang,Zhuliang Yu,Zhenghui Gu,Siyuan Chen
标识
DOI:10.1109/tii.2026.3728132
摘要
To address the infeasibility issue of conventional fault-tolerant control methods for redundant manipulators under joint lockup, this article proposes a fault-tolerant recurrent neural network based on a feasible space expansion strategy fault-tolerant recurrent neural network based on a feasible space expansion strategy (FSES-FTRNN). By introducing a virtual scaling variable, the proposed method dynamically reconstructs task constraints, transforming the original infeasible optimization problem into a feasible one while strictly preserving safety constraints. Compared with traditional approaches that rely solely on residual redundancy and fail when constraint spaces become disjoint, the proposed method preserves the intended geometric path through adaptive constraint-space expansion and online execution-rate scaling. Experiments on a six-degree-of-freedom (6-DOF) Kinova Jaco$^{2}$manipulator under progressive multijoint lockup scenarios demonstrate complete geometric-path execution in cases where conventional quadratic programming methods lose feasibility, while maintaining high path-tracking accuracy under varying constraint intensities. These results indicate that the proposed approach provides a practical velocity-level fault-tolerant command-generation scheme for redundant manipulators under high-impact joint-lockup conditions.
科研通智能强力驱动
Strongly Powered by AbleSci AI