Learning-Based Adaptive Optimal Control for Dual-Arm Unmanned Aerial Manipulator Systems Catching Unknown Objects With Safety Constraints

Dual-arm aerial manipulators enable coordinated and dexterous operations. However, the strong dynamic coupling among the vehicle, the two arms, and the grasped object complicates stable control under safety constraints. This article develops a comprehensive kinematic-dynamic model that explicitly incorporates the manipulated object and proposes an adaptive dynamic programming controller with a critic–actor neural network. A control barrier function is integrated to enforce asymmetric state and safety limits. The proposed controller is capable of managing unknown dynamics and actuator saturation while simultaneously balancingoptimality, safety, robustness,andmanipulability.Hardware experiments on a self-built dual-arm aerial platform validate the approach on trajectory-tracking tasks with grasped payloads and wind-like perturbations. The proposed controller achieves accurate tracking, maintains stability, and respects safety constraints across operating conditions, outperforming conventional controllers in both tracking accuracy and robustness, thereby supporting practical aerial manipulation.