On underwater robot manipulators, high speed and high precision are basic requirements in order to improve efficiency of operations. To satisfy these requirements, feedforward control inputs are crucial. For making feedforward inputs, one method is to estimate all parameters of the robot dynamics including hydrodynamic terms such as added-mass, drag force, and buoyancy. However, the parameter estimation of hydrodynamic coefficients is not suitable for forming the feedforward control inputs of underwater robot manipulators because it is difficult to model and estimate the hydrodynamic terms. To overcome such a difficulty, we apply iterative learning control to underwater robots. In this paper, we theoretically and experimentally investigate the performance of iterative learning control for underwater robot manipulators.
Underwater human operations involve high risks because currents, waves, water pressure and fluid resistance make the operations dangerous, hard and inefficient. For these reasons, many underwater robot vehicles have been studied and developed (e.g. Fossen (1994), Yoerger (1985), George Lee (2001)). Furthermore, underwater manipulators mounted on the vehicles that work in the water instead of human arms have been studied and many papers on manipulators have been published (e.g. Scholberg (1994), Kato (1996), Shintaku (1999), Canudas-de-Wit (2000), Chung (2001), Sarkar (2001)). In these papers, they confirmed the validity of their works through the simulation results with some assumptions not through experimental results. The main reason is that the actual dynamics of underwater robot manipulators with multi-degree of freedom is much complicated. For example, if we consider three-dimensional water flows into dynamics, it is extremely difficult to treat the direction and the magnitude of water flows that act on each part of the links of a manipulator. In other wards, most of the previous works ignored the hydrodynamic factors which cannot be easily modeled.