Dynamics　control　of　free-floating　space　robot　systems　with　joint　torque　output　dead-zones,　friction　and　external　disturbances　were　discussed.　A　recurrent　CMAC　and　an　adaptive　dead-zone　compensator　were　designed　to　track　the　desired　trajectories　of　two　joints.　Friction　double　observers　were　designed　to　estimate　the　unmeasurable　internal　friction　states,　and　dead-zone　precompensator　was　designed　to　eliminate　the　impacts　of　joint　torque　output　dead-zones.　The　recurrent　cerebellar　neural　network　model　was　applied　to　approximate　uncertainty　terms　of　dynamics　equation　including　friction　errors　and　external　disturbances.　The　results　of　simulation　show　the　efficiency　of　the　proposed　control　method.　©　2018,　China　Mechanical　Engineering　Magazine　Office.　All　right　reserved.