Dynamic Self-Triggered Model Predictive Control for Modular and Reconfigurable Robots via Data-Driven Adaptive Dynamic Programming

The modular and reconfigurable robotic system (MRRS) is increasingly deployed in highly challenging operational environments—such as deep-sea exploration and autonomous agriculture—where communication bandwidth, onboard computational resources, and energy supply are severely constrained, rendering direct human supervision or intervention impractical. Addressing this challenge, we propose a novel control framework that synergistically integrates dynamic self-triggered model predictive control (DSTMPC) with data-driven adaptive dynamic programming (ADP), specifically tailored for MRRS operating under uncertain dynamics and input constraints. First, the Newton–Euler method is employed to formulate the system dynamics model, and a data-driven predictive model is developed using a recurrent neural network (RNN) to reconstruct the unknown dynamic components. Within the ADP framework, a time-varying Hamilton–Jacobi–Bellman (HJB) equation is formulated, and the optimal control policy is approximated via an actor–critic neural network (NN) architecture. A novel dynamic self-triggering mechanism (DSTM), based on both actual and predicted tracking errors, is designed to avoid Zeno behavior. Rigorous Lyapunov-based analysis proves that all error variables are uniformly ultimately bounded (UUB). Finally, the feasibility of the proposed algorithm is validated on a 2-degree-of-freedom (DOF) modular and reconfigurable robot (MRR) experimental platform. Experimental results demonstrate strong robustness and promising application potential in complex scenarios.