This paper presents gain optimization for a controller of a 6- DOF underwater robot with tilting thrusters. PID control system with anti-windup technique is designed to stabilize the hovering motion of the robot. The controller comprises thrust vector decomposition to overcome nonlinearity of the thrust vector and also includes an algorithm to compensate for saturation of thrusters. A total of 24 control gains should be tuned in this controller, and gain optimization is performed according to four system errors using genetic algorithm. First, 18 PID control gains were optimized and then 6 gains were optimized to affect anti-windup. As a result, control gains optimized by the integral absolute error showed the best performance, and it is verified that tracking error in position and orientation of the robot were reduced by 29.38% compared with initial gains.