An Adaptive Data-Driven Controller For Underwater Manipulators With Variable Payload

The underwater environment poses a complex problem for developing control systems for underwater manipulators. Modeling the system is a complicated and costly process due to the highly nonlinear dynamics and the presence of unknown hydrodynamical effects. Furthermore, manipulators are usually deployed on Autonomous Underwater Vehicles (AUVs) which further influence and change the dynamics of the arm. This is aggravated in underwater operations where manipulating different objects is necessary. These diverse manipulation tasks introduce external disturbances to the system and can lead to a fast degradation of the control system performance. In this article, we propose a novel adaptive controller for underwater robot manipulators that have to handle varying payloads with different masses, geometries, and buoyant forces. The proposed control strategy utilizes a data-driven model of the system in an optimal control formulation based on neural networks. Moreover, we developed an online tuning strategy, based on the adaptive interaction theory, which allows the gains of the controller to be updated online with respect to a set of performance metrics. Experiments were performed with the robotic arm manipulating a variety of payloads while mounted on both a fixed base and a free-floating vehicle. We present a number of simulated and experimental results that illustrate the benefits of the proposed strategy. In addition, a comparative study against a classical Model Predictive Control (MPC) demonstrates the benefits of our adaptive proposal.