A Probabilistic Framework For Hydrodynamic Parameter Estimation for Underwater Manipulators

This paper presents an approach for the dynamic model parameter estimation for underwater manipulators. Accu-rate dynamic modeling of underwater robotic manipulators is im-portant for the design of high-performance control architectures and the creation of high-fidelity simulators. In this paper a two-step approach combining a least-square method with a Monte Carlo Markov Chain (MCMC) approach is presented based on experimental data gathered with a 4-degree-of-freedom (DOF) manipulator. The parameters estimated are the manipulator links inertia, the center of mass for the links, the drag coefficients, and the friction coefficients. A detailed investigation into the importance of the prior distributions of these parameters is presented. The proposed strategy is evaluated by looking at the posterior distributions as well as by analyzing the effects of these parameters in the dynamic model in comparison to the real behavior of the system.