Author(s)

GuangYu Long, ZhuoSong Fu^*

Affiliation(s)

College of Computer Science and Engineering, Jishou University, Jishou 416000, Hunan, China.

Corresponding Author

ZhuoSong Fu

ABSTRACT

Zeroing Neural Network (ZNN) is widely used for robotic trajectory tracking due to their rapid convergence and strong error attenuation properties. In discrete implementations, however, the choice of step-size critically influences tracking accuracy, numerical stability, and computational cost. Conventional variable-step strategies typically rely on fixed heuristics or manually tuned rules, limiting their adaptability in dynamic task conditions. To address this limitation, this paper introduces a Large Language Model (LLM)-based adaptive step-size mechanism for discrete ZNN in manipulator trajectory tracking. The LLM receives natural-language instructions together with essential system-state descriptions and outputs step-size adjustments that guide the ZNN update. This enables intuitive human-robot interaction and allows the controller to flexibly shift between high-precision tracking and low-computation execution without modifying the underlying ZNN formulation. Results show that the proposed method improves tracking accuracy when finer steps are selected, reduces computational load when coarser steps suffice, and maintains high semantic consistency in interpreting step-size-related instructions. These findings demonstrate the potential of integrating LLM reasoning into step-size regulation to enhance the flexibility and interpretability of discrete ZNN-based robotic tracking.

KEYWORDS

Zeroing neural network; Variable step-size; Large Language Model; Robotic manipulator

CITE THIS PAPER

GuangYu Long, ZhuoSong Fu. LLM-guided adaptive step-size zeroing neural network for robotic manipulator trajectory tracking. World Journal of Engineering Research. 2025, 3(6): 1-6. DOI: https://doi.org/10.61784/wjer3064.

REFERENCES

[1] Fu Z, Zhang Y, Tan N. Gradient-feedback Zhang neural network for unconstrained time-variant convex optimization and robot manipulator application. IEEE Transactions on Industrial Informatics, 2023, 19(10): 10489-10500.

[2] Fu Z, Zhang Y, Li W. Solving future nonlinear equation system via ZNN and novel general ILR3S formula with multitype manipulator applications. IEEE Transactions on Industrial Electronics, 2024, 71(10): 12623-12633.

[3] Lai G, Lin C, Ouyang Y, et al. Finite-time uncalibrated visual servoing for robotic manipulators based on model-free zeroing neural networks. IEEE Transactions on Automation Science and Engineering, 2025, 22: 24347-24361.

[4] Li J, Qu L, Zhu Y, et al. A novel zeroing neural network for time-varying matrix pseudoinversion in the presence of linear noises. Tsinghua Science and Technology, 2025, 30(5): 1911-1926.

[5] Liao B, Han L, Cao X, et al. Double integral-enhanced Zeroing neural network with linear noise rejection for time-varying matrix inverse. CAAI Transactions on Intelligence Technology. 2024, 9(1): 197–210.

[6] Guo J, Zhang Y. Stepsize interval confirmation of general four-step DTZN algorithm illustrated with future quadratic programming and tracking control of manipulators. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2021, 51(3): 1662-1670.

[7] Qiu B, Guo J, Li X, et al. New discretized zeroing neural network models for solving future system of bounded inequalities and nonlinear equations aided with general explicit linear four-Step rule. IEEE Transactions on Industrial Informatics, 2021, 17(8): 5164-5174.

[8] Park J, Park C, Park P. A variable step-size MSD-based diffusion LMS algorithm with nnoisy input and quantization link noise. IEEE Wireless Communications Letters, 2025, 14(11): 3734-3738.

[9] Liu J. An improved arithmetic optimization algorithm with multi-strategy variable step size for optimizing engineering problems. IEEE Access, 2025, 13: 156566-156599.

[10] Shen L, Jiang D, Zhu H. Control study for suppressing rotor vibration of the six-pole hybrid magnetic bearing system based on no-parameter variable-step-size least-mean-square algorithm. IEEE Transactions on Power Electronics, 2025, 40(8): 10672-10681.

[11] Hongying Z, Javed A, Abdullah M, et al. Large language models with contrastive decoding algorithm for hallucination mitigation in low-resource languages. CAAI Transactions on Intelligence Technology. 2025, 10: 1104-1117.

[12] Duan J, Li W, Bai Q, Nguyen M, et al. LLM-botguard: A novel framework for detecting LLM-driven bots with mixture of experts and graph neural networks. IEEE Transactions on Computational Social Systems, 2025, 12(5):  3488-3500.

[13] Chen W, Li G, Li M, et al. LLM-enabled incremental learning framework for hand exoskeleton control. IEEE Transactions on Automation Science and Engineering, 2025, 22: 2617-2626.

[14] Fakhoury S, Naik A, Sakkas G, et al. LLM-based test-driven interactive code generation: User study and empirical evaluation. IEEE Transactions on Software Engineering, 2024, 50(9): 2254-2268.

[15] Ferro M, Mirante A, Ficuciello F, et al. A Coppeliasim dynamic simulator for the Da Vinci research kit. IEEE Robotics and Automation Letters, 2023, 8(1): 129-136.