An　adaptive　decentralized　neural　network　fault-tolerant　algorithm　based　on　state　observer　is　proposed　for　a　dual-arm　space　robot　system　under　partial　loss　of　joint　actuator　effectiveness.　The　dynamic　equation　of　the　space　robot　system　is　established　based　on　Lagrange＇s　second　method.　The　problem　of　fault　tolerance　of　actuator　failure　is　transformed　into　the　one　of　adaptive　control　of　nonlinear　interconnected　system　with　uncertain　parameters　by　utilizing　the　decentralization　theory.　The　angular　velocity　signals　of　the　system　are　obtained　by　using　the　state　observer,　and　the　uncertain　terms　and　the　interconnected　terms　are　estimated　by　the　adaptive　decentralized　neural　networks.　The　stability　criterion　of　controller　and　observer　are　given　based　on　the　Lyapunov　function　method.　The　numerical　simulation　shows　that　high-precision　trajectory　tracking　control　can　be　realized　by　the　controller　within　2s,　and　the　actual　angular　velocity　signals　of　joints　can　be　estimated　accurately　by　the　state　observer　regardless　of　whether　the　actuator　faults　occur　or　not,　hence　the　correctness　of　the　theoretical　analysis　and　the　feasibility　of　the　algorithm　are　verified.　©　2019,　Editorial　Department　of　Journal　of　Chinese　Inertial　Technology.　All　right　reserved.