Trajectory Tracking Closed-Loop Cooperative Control of Manipulator Neural Network and Terminal Sliding Model

Deqing Liu, Zhonggang Xiong, Zhong Liu, Mengyi Li, Shunjie Zhou, Jiabao Li, Xintao Liu, Xingyu ZhouSymmetry 2025 · DOI

TL;DR

This paper tackles the challenge of achieving high-accuracy trajectory tracking in six-degree-of-freedom robotic arms. The authors propose a novel control strategy that combines a Radial Basis Function Neural Network (RBFNN) for approximating uncertainties with a non-singular fast terminal sliding mode (NFTSM) controller to ensure rapid and stable convergence. Their approach demonstrates superior tracking performance compared to existing methods in both simulations and real-world experiments.

What Problem Does This Paper Address?

Robotic arms, particularly those with multiple degrees of freedom (like the six-DOF manipulator discussed), are crucial in various industries. However, precisely controlling their movement to follow complex trajectories is a significant challenge. This difficulty arises from inherent system complexities: the manipulator's dynamics are highly nonlinear and coupled, and they are further complicated by model errors and parameter uncertainties. Traditional control methods often struggle with slow tracking speeds, low accuracy, and issues like "chattering" (unwanted oscillations) or singularity problems. Existing advanced techniques, while improving accuracy, can suffer from high computational complexity, limiting their real-time applicability. This paper aims to bridge this gap by developing a control strategy that is both highly accurate and computationally efficient.

Methods

The core of this paper lies in its hybrid control strategy:

  • Dynamic Model: The authors first derive the dynamic model of the six-DOF manipulator using the Lagrangian method. This forms the foundation for understanding the robot's behavior.
  • Non-Singular Fast Terminal Sliding Mode (NFTSM) Control: To address the limitations of traditional sliding mode control, they introduce an NFTSM surface. This design aims to accelerate convergence to the desired trajectory and, crucially, avoid singularity issues. It achieves this by incorporating nonlinear and fast terminal terms that dominate when the error is large, quickly pulling the system towards the equilibrium, and linear terms that ensure precise convergence when the error is small.
  • Radial Basis Function Neural Network (RBFNN) Compensation: To handle the unknown and varying uncertainties in the manipulator's model, an RBFNN is employed. This network acts as an adaptive compensator, learning and approximating these uncertainties online. The RBFNN is designed with a specific structure (input, hidden, output layers) and uses Gaussian basis functions in its hidden layer. Its weights are adaptively updated to minimize the approximation error.
  • Lyapunov Stability Theory: The stability of the entire closed-loop control system is rigorously proven using Lyapunov stability theory. A Lyapunov function is constructed, and its time derivative is shown to be negative definite, guaranteeing that the system will converge to a stable state.
  • Cooperative Control Law: The final control input is a combination of the RBFNN's approximation of the equivalent control law and the switching control law from the NFTSM controller. This cooperative approach leverages the strengths of both components.

Datasets

Not discussed in the paper. The evaluation relies on simulations within MATLAB/Simulink and experimental validation on a JAKA Zu5 robotic arm.

Key Insights and Ideas

  • Synergy of RBFNN and NFTSM: The paper highlights that combining an adaptive neural network for uncertainty compensation with a robust sliding mode controller designed to avoid singularities and accelerate convergence leads to superior performance. The RBFNN handles the "unknowns," while the NFTSM ensures fast and stable tracking.
  • Addressing Singularity in Sliding Mode: The design of the non-singular fast terminal sliding surface is a key innovation. It effectively circumvents the singularity problem often encountered in traditional sliding mode control, which is critical for real-world robotic applications.
  • Computational Efficiency: While not explicitly quantified in terms of FLOPS, the paper implies that the RBFNN's structure and the NFTSM's design contribute to a practical and computationally feasible solution, contrasting with some prior methods that suffered from high complexity.
  • Experimental Validation: The successful validation on a physical JAKA Zu5 robot arm is a strong indicator of the method's practical applicability beyond simulation.

Key Results

The proposed RBFNN-NFTSM control method demonstrated significant improvements in trajectory tracking accuracy:

  • Position Tracking: In simulations, the proposed method achieved a maximum tracking error of approximately 0.05 rad, outperforming RBFNN-SMC (0.1 rad) and NFTSM (0.15 rad). The mean absolute errors for the first three joints were 0.03 rad for RBFNN-NFTSM, 0.04 rad for RBFNN-SMC, and 0.05 rad for NFTSM.
  • Velocity Tracking: Experimental results showed that the proposed controller achieved a significantly reduced angular velocity error of only 0.03 rad/s, compared to 0.5 rad/s for RBFNN-SMC and 0.18 rad/s for NFTSM.
  • Smoother Responses: Both simulation and experimental results indicated that the proposed method produced smoother response curves with less fluctuation, indicating better control robustness and stability.

Limitations and Future Work

  • Controlled Environment: The authors acknowledge that the experimental validation was conducted in a controlled laboratory environment. They note that real-world industrial scenarios may present more complex external factors that could affect the manipulator's control robustness.
  • Future Work: The authors plan to conduct further validations in environments that more closely resemble real-world operating conditions to assess the controller's performance under more challenging and unpredictable circumstances.

Summary updated

BibTeX
@article{liu2025trajectory,
  author = {Deqing Liu and Zhonggang Xiong and Zhong Liu and Mengyi Li and Shunjie Zhou and Jiabao Li and Xintao Liu and Xingyu Zhou},
  title = {{Trajectory Tracking Closed-Loop Cooperative Control of Manipulator Neural Network and Terminal Sliding Model}},
  journal = {Symmetry},
  year = {2025},
  doi = {10.3390/sym17081319},
}
Figure