Online system identification of the dynamics of an Autonomous Underwater vehicle

Abstract

Autonomous Underwater vehicles (AUV) with reconfigurable payloads are rapidly becoming common. Their dynamic characteristics are affected when payloads change. Typically, retuning of the controller is required to maintain good control performance. To address this situation, we develop a technique to enable rapid identification of AUV dynamics online. We demonstrate the technique with a fin-controlled single-thruster torpedo-shaped AUV. By decoupling the system according to planar and horizontal motion, mathematical models for yaw and pitch dynamics are developed. This results in a second-order transfer function with auxiliary steady state fin deflection. Identification of continuous-time model was performed to preserve the physical meaning of the parameters. Identification in continuous-time requires time-derivative terms which are reconstructed using the state variable filter (SVF). Then, recursive least-square (RLS) algorithm is used to identify the unknown parameters. The proposed identification method was validated through field deployments of our AUVs. The online estimates compare favorably with results obtained from offline identification methods requiring numerical optimization. We demonstrate how turning radius of the AUV can be estimated accurately from the identified parameters. We also show how a gain-scheduled controller, with better control performance than a constant-gain controller, can be designed using the estimated parameters.