Robot　navigation　is　important　symbol　of　its　intelligent　level,　especially　under　dynamic　environment.　In　this　paper,　a　robot　navigation　method　based　on　intelligent　evolution　is　proposed,　and　Tolman　Mouse　Maze　Experiment　is　reproduced　by　robot　rather　than　mouse.　Firstly　the　topological　map　of　the　indoor　environment　is　built　by　Self-Organizing　Map　(SOM)　method　based　on　random　traversal,　then　the　path　planning　method　based　on　reinforcement　learning,　during　the　navigation　process,　the　environment　of　the　maze　will　change,　and　the　robot　can　automatically　update　its　topological　map　according　to　the　environment　change　by　adjusting　its　neural　network　structure,　so　as　to　reach　the　goal　as　soon　as　possible.　The　validity　of　the　proposed　method　is　proved　by　simulation　results　based　on　MATLAB　and　gazebo,　moreover　physical　experiments　carried　on　Turtlebot　robot　both　reproduced　Tolman　Mouse　Maze　Experiment,　so　our　intelligent　evolution　algorithm　made　robot　as　clever　as　animal　being　able　to　find　the　short　path　to　the　goal.　©　2018　IEEE.