TIE MPC

Traditional hierarchical remotely operated vehicle (ROV) control suffers from feasibility gaps between motion control and thrust allocation (TA). Modeling uncertainties further complicate the control problem. While GP-MPC can handle these uncertainties, it introduces non-convex optimization with prohibitive computational costs.This paper proposes a sparse linearized gaussian process-based direct thruster optimization MPC strategy (SL-GP-MPC) that addresses both control architecture and modeling problems. The proposed approach eliminates the TA layer by formulating thruster forces as direct optimization variables, thereby removing the feasibility gap inherent in hierarchical control structures. Furthermore, sparse linearized Gaussian processes enable convex approximation of the non-convex GP-MPC, significantly improving computational efficiency while preserving complete probabilistic modeling capabilities. Theoretical analysis demonstrates the algorithm's recursive feasibility and asymptotic stability. Pool experiments validate the superior performance of the proposed method in complex hydrodynamic environments.