Efficient　navigation　in　complex　environment　is　one　of　research　hotspots　in　the　field　of　robot　control.　In　this　paper,　for　the　problems　of　navigation　in　distributed　environment　of　a　mobile　robot,　we　propose　a　regionalization　navigation　method　based　on　deep　reinforcement　learning.　First　of　all,　consider　the　characteristics　of　distributed　environment,　we　use　independent　submodules　learn　control　strategy　in　different　region,　and　region　model　is　built　to　integrate　strategies　to　complete　navigation　in　multi-area　environment.　Then,　in　order　to　improve　learning　efficiency,　reward　prediction　and　depth　obstacles　avoidance　are　added　during　training.　Experiment　result　in　single-area　reveal　the　improvements　of　training　method　is　helpful　to　enhance　robot　navigation　performance.　Moreover,　by　the　proposed　regionalization　navigation　studying　in　multi-area　environment,　our　method　shows　the　advantages　in　training　time　and　reward　that　single　model　does　not　have,　indicate　that　it　can　better　deal　with　large-scale　navigation.　©　2019　IEEE.