The　variable　time　delay　and　the　packet　loss　in　the　Internet　degrade　seriously　the　performance　of　Internet　based　teleoperation　system,　and　even　make　the　system　unstable.　To　overcome　this　problem,　a　novel　control　architecture　of　the　Internet-based　teleoperation　system　is　proposed.　The　reference　information　signed　with　time　stamp　at　master　site　is　transmitted　to　the　slave　site　and　then　feeds　back　to　the　operator,　so　the　operator　can　get　the　round-trip-timedelay　(RTT)　and　then　predict　the　next　RTT　making　use　of　the　multivariate　linear　regression　algorithm　to　analyze　the　preceding　RTT　which　has　been　got.　A　generalized　predictive　control　(GPC)　strategy　using　CARIMA　model　of　the　lineared　slave　robot　is　designed　at　slave　site　to　control　the　telerobot.　Consequently,　the　operation　performance　of　the　system　is　improved　by　using　this　method.　Simultaneously,　the　GPC　controller　generates　the　redundant　control　information　which　can　diminish　the　influence　of　the　packet　loss　and　the　large　time　delay　in　the　Internet　to　the　system.　Finally,　stability　condition　is　deduced　according　to　the　predictive　stability　theorem.　Simulation　results　show　that　these　strategies　can　solve　efficiently　the　problem　caused　by　the　variable　time　delay　and　the　packet　loss.