.png){kind=icon}
.png){kind=icon}
Return
A Predefined-Time Convergent Dual-Channel Fuzzy Attention RNN for Motion Planning of Robotic Systems: Application to Robot-Assisted Puncture
.png){kind=icon}
PRE
AI
B
C
D
Z
J
P
DOI：10.1109/TFUZZ.2026.3684056.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Although recurrent neural networks (RNNs) have demonstrated remarkable performance in robotic motion planning, most existing RNN-based approaches still impose uniform and identical gain parameters across all network dimensions to regulate convergence behavior. This practice ignores the fact that error magnitudes and dynamic characteristics can vary significantly across different dimensions, which may lead to suboptimal convergence performance, reduced control precision, and even instability in certain cases. To overcome this limitation, we propose a novel predefined-time convergent dual-channel fuzzy attention RNN (PTC-DCFA-RNN) for robotic motion planning. Unlike conventional RNNs, the proposed PTC-DCFA-RNN incorporates two key innovations: first, a dual-channel fuzzy attention mechanism that adaptively and in real time regulates the convergence behavior of each dimension by assigning distinct gain parameters based on its error magnitude and dynamic characteristics, and second, a fuzzy underlying acceleration activation function specifically designed to efficiently enhance convergence speed. Theoretical analysis establishes the predefined-time convergence of the state trajectory of the PTC-DCFA-RNN to a Karush–Kuhn–Tucker consistent solution. Extensive path-tracking simulations and real-world robot-assisted puncture experiments show that the proposed method achieves faster convergence and higher position and orientation accuracy than state-of-the-art RNN-based approaches. In summary, our method opens a new perspective on the critical problem of gain regulation in RNN-based methods and may inspire a new line of research on adaptive, dimension-wise scheduling for predictable convergence.
Keywords:
Recurrent neural network (RNN)
robotic system
time-varying quadratic programming (TVQP)
Journal
IF:
11.9
Papers:
4.9K
Citations:
2.9W
Organization
Cited Papers