Insights into a data driven optimal control for energy efficient manipulation

To enable underwater manipulators with long-lasting autonomy, designing an energy efficient controller is of utmost importance. In this regard, an optimal control technique is a suitable approach as the cost function allows optimization for different metrics, such as energy consumption, minimum velocity changes, or zero position errors. However, the need for an accurate model makes optimal control strategies less enticing for underwater systems where models are difficult to obtain due to unknown dynamics. A solution for this limitation is the usage of data driven techniques for model prediction, as they solely rely on the observed behaviour of the system for generating dynamic models. In this paper, we study the capabilities of a data driven model predictive controller for energy-efficient underwater manipulation tasks. A data driven model of the underwater manipulator based on a neural network is integrated into the formulation of a well known Model Predictive Control (MPC). The proposed architecture is implemented on a four Degrees-of-Freedom (DOF) underwater manipulator in a simulated environment and the results are presented in comparison with a classical MPC controller, showcasing the benefits of the proposed data driven strategy.