Particle　impoverishment　is　inevitably　induced　because　of　the　random　particles　prediction　and　resampling　applied　in　particle　filter,　especially　in　SLAM　problem　with　a　large　number　of　dimensions.　To　overcome　these　limitations,　an　improved　FastSLAM　system　using　artificial　fish-swarm　optimized　distributed　unscented　particle　filter　(AFSO-DUPF)　is　developed　in　this　paper.　The　main　contribution　of　this　paper　is　the　derivation　of　AFSO-DUPF　by　improving　the　resampling　method　and　reduce　the　computation　quantity　of　distributed　unscented　particle　filter　(DUPF).　First,　the　intelligent　optimized　resampling　method　basing　on　artificial　fish-swarm　algorithm　was　used　to　overcome　degeneracy　and　losing　particle　density　simultaneously.　Second,　to　overcome　the　higher　computational　complexity　problem,　the　marginalized　distribution　simplified　method　was　used　in　sampling　particles　and　calculating　weights　of　DUPF.　Compare　to　DUPF　FastSLAM,　the　AFSO-DUPF　FastSLAM　can　maintain　the　diversity　of　particle　and　avoid　inconsistency　for　long　time　periods.　The　simulation　experiment　results　demonstrate　that　the　improved　FastSLAM　system　has　higher　estimation　accuracy　and　reduce　the　computational　complexity.　©　2017　Technical　Committee　on　Control　Theory,　CAA.