A　decentralized　fault-tolerant　control　method　was　designed　based　on　non-singular　terminal　sliding　mode　for　a　space　robots　subjected　to　partial　losses　of　actuator　effectiveness.　Firstly,　the　dynamics　equation　of　the　system　was　established　by　using　linear　momentum　conservation　law　and　Lagrangian　method.　Subsequently,　the　system　was　decentralized　into　several　subsystems　based　on　the　local　informations　of　the　carrier　and　joints,　and　the　dynamics　equations　of　subsystems　were　obtained.　Then　the　effective　factors　were　separated　from　the　subsystems＇　dynamic　equations,　and　the　adaptive　decentralized　neural　network　was　used　to　estimate　the　separated　variables　in　real　time.　According　to　the　estimation　results,　the　corresponding　control　law　was　designed　online　to　counteract　the　influences　of　the　actuator　faults　on　the　stability　of　system　and　ensure　good　trajectory　tracking　performance.　The　asymptotic　stability　of　the　whole　closed-loop　system　may　be　guaranteed　under　the　proposed　control　scheme　via　Lyapunov　function　method.　The　simulation　results　demonstrate　the　effectiveness　of　the　proposed　control　method.　©　2019,　China　Mechanical　Engineering　Magazine　Office.　All　right　reserved.