Text this: Optimal control of manipulator with joint clearance compensation via generalized policy learning