Experimental study on a learning control system with bound estimation for underwater robots

Abstract

Underwater robotic vehicles (URVs) have become an important tool for numerous underwater tasks due to their greater speed, endurance, depth capability, and a higher factor of safety than human divers. However, most vehicle control system designs are based on simplified vehicle models and often result in poor vehicle performance due to the nonlinear and time-varying vehicle dynamics having parameter uncertainties. Conventional proportional-integral-derivative (PID) type controllers cannot provide good performance without fine-tuning the controller gains and may fail for sudden changes in the vehicle dynamics and its environment. Conventional adaptive control systems based on parameter adaptation techniques also fail in the presence of unmodeled dynamics. This paper describes a new vehicle control system using the bound estimation techniques, capable of learning, and adapting to changes in the vehicle dynamics and parameters. The control system was extensively “wet-tested” on the Omni-Directional Intelligent Navigator (ODIN)-a six degree-of-freedom, experimental underwater vehicle developed at the Autonomous Systems Laboratory, and its performance was compared with the performance of a conventional linear control system. The results showed the controller's ability to provide good performance in the presence of unpredictable changes in the vehicle dynamics and its environment, and it's capabilities of learning and adapting.