Abstract: Traditional singular perturbation apprbach can only be used in robot systems with weak joint flexibility.In order to solve this problem,a joint flexibility compensator is designed,which can greatly increase the equivalent joint stiff- ness,so the limitation of joint flexibility for singular perturbation approach is eliminated and the approach can be extended to the robot system with normal joint flexibility.Besides,an adaptive controller with projection algorithm as parameter estimation law is designed for the slow subsystem control and its asymptotical stability is proved.The control strategy can be applied conveniently in engineering without any joint flexibility limitation and does not need link acceleration or its differential signal.Finally,an experimental study on the flexible joint manipulator is executed and the results verify the validity and feasibility of the proposed strategy.