Type-2 Fuzzy Single Hidden Layer Recurrent Neural Adaptive Terminal Super-Twisting Control of Robot Joint

This research proposes a neural network-based super-twisting controller for robot joints. A modified fast nonsingular terminal sliding surface is introduced, which not only avoids singularity but also increases the convergence rate of the sliding mode control. To address the challenge of system uncertainty modeling, a type-2 fuzzy single hidden layer recurrent neural network (T2FSHLRNN) is proposed. The T2FSHLRNN, configured as a weighted combination of a type-2 fuzzy neural network and a single hidden layer network, demonstrates strong global learning ability. Leveraging its internal and external double-layer feedback mechanism, the network can incorporate both current and previous error information during the approximation process, effectively improving the approximation accuracy and reducing system chattering. Furthermore, an adaptive gain function is proposed and an adaptive terminal super-twisting controller based on T2FSHLRNN (ATSC-T2FSHLRNN) is developed. The system’s stability under unknown disturbance is ensured using Lyapunov synthesis. Based on this, the online parameter learning algorithm for T2FSHLRNN and the variable gains of ATSC are derived. Simulation confirms the effectiveness of the proposed ATSC-T2FSHLRNN.