Sliding mode control of robot manipulators via an improved recurrent neural network and barrier functions

Robotic manipulators face challenges in trajectory tracking due to uncertainties and external disturbances. Existing methods often need accurate models, multiple adaptive networks, or fixed switching gains. Researchers have designed a new adaptive terminal sliding-mode controller that uses a dual-feedback Improved-RNN and a super-twisting mechanism with barrier functions.

This controller uses a single Improved-RNN to approximate the lumped equivalent control term, while the switching gain transitions from a time-varying reaching gain to a state-dependent barrier-function gain. By applying Lyapunov analysis, the closed-loop system is guaranteed to converge in finite time. The proposed controller was compared to existing fixed-time controllers and an RBFNN in identical nominal and sudden payload-change situations.

The new method showed the lowest nominal tracking root mean square error and shortest nominal settling times for both joints, and maintained small tracking errors after a sudden payload change without requiring controller retuning. This research was supported by Dezhou Intelligent Equipment Research and Development Center under Grant PT2025KJT004, and the authors contributed equally.

The study is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.