Abstract: For the underwater cleaning of water conveyance structures, a tracked underwater cleaning robot is developed, including a crawler, a five-function electric manipulator and an electric rotary brush. A hybrid force/position controller for underwater cleaning is proposed, consisting of a force controller and a trajectory tracking controller. The force controller adopts an adaptive PID based on neural network(NNPID) to control the vertical contact force on the work surface. The trajectory tracking controller adopts a robust model predictive control(RMPC) based on an extended state observer(ESO), which can effectively compensate for internal model errors and underwater environmental disturbances, achieving end-effector trajectory tracking in the horizontal direction of the work surface. Onshore tests and actual underwater engineering experiments demonstrate that the robot can perform underwater cleaning operation and effectively clean surface dirt along the desired trajectory with the end-effector contact force errors of ±1 N and the trajectory tracking errors of ±2 mm.