Model predictive control based on generalized fuzzy hyperbolic model applied to modular and reconfigurable robots with full-state constraints via adaptive dynamic programming

Modular and reconfigurable robots (MRRs) hold significant potential for deployment in complex and constrained environments, such as underwater reconnaissance, aerospace operations, and confined space tasks, where ensuring safety and performance under full-state constraints is critical. To address the challenges of unknown dynamic terms and full-state constraints in MRRs, this paper proposes an model predictive control (MPC) strategy based on adaptive dynamic programming (ADP) with a generalized fuzzy hyperbolic model (GFHM). First, the Newton-Euler iterative method is employed for dynamic modeling, and an identifier based on the GFHM is developed to reconstruct unknown dynamic terms. Subsequently, the MPC strategy based on ADP is implemented to construct an actor-critic neural network (NN) architecture incorporating time-varying activation functions. A barrier-type cost function is introduced to guarantee constraint satisfaction and forward invariance of the safety set. Within a backstepping control structure, both virtual and actual optimal controllers are designed. Lyapunov analysis rigorously proves that all error variables are uniformly ultimately bounded (UUB). Experimental validation on a 6-DOF Sawyer MRR platform demonstrates the effectiveness of the proposed approach, which achieves stable trajectory tracking while respecting all state constraints, indicating its applicability in real-world scenarios requiring safe and adaptive robot operation.