This paper describes a method to design adaptive attitude controller of autonomous underwater vehicle (AUV). The main purpose of this design is to adapt to the error caused by coupling between different controls channels as time-varying errors. In this paper, system identification method is not used to get the dynamic model directly, but to simplify the hydrodynamic model and quantize the effect caused by coupling as error of parameters. Computational-fluid-dynamics (CFD) method is used to build a hydrodynamic model to increase the experiment data and simulate the effect of coupling. And on this basis, An ℒ1 adaptive controller can get enough resources to be build. In the final simulation, the acquired controller reveals better performance than an elaborately tuned PID controller. The ℒ1 adaptive controller can handle the coupling better. The output of the controlled system can follow a response of a given transfer function, which benefits the controller design in outer loop.