Abstract
Ultrasonic thickness inspection of ship bulkheads poses significant challenges due to confined spaces, dynamic obstacles, and highly variable environments. This paper presents a novel autonomous robotic arm control framework tailored for such conditions, combining enhanced Unscented Kalman Filter (UKF) with a hierarchical Model Predictive Control (MPC) strategy. We introduce a residual-driven adaptive noise covariance UKF (RD-ANC) integrated with a Huber penalty function (HP-UKF), significantly improving robustness against sensor noise and outliers during real-time mapping and estimation. A Three-Layer Energy-Efficient MPC (TLE-MPC) is designed, comprising: a global planner using Differential Dynamic Programming (DDP) for energy budgeting and coarse path generation; a coordination layer using Sequential Quadratic Programming (SQP) for obstacle avoidance and adaptive energy trade-offs; and an execution layer leveraging Explicit MPC (eMPC) for sub-5 ms control law computation. Simulation results show the framework achieves real-time obstacle avoidance, stable path tracking, and up to 15% energy reduction during inspection tasks in semi-structured and unpredictable ship environments. This research offers a robust and scalable method for autonomous robotic inspection and lays the foundation for future multi-arm cooperation and long-duration energy-aware deployments.