Spatial　cells　in　hippocampus　play　a　functional　role　in　representation　and　processing　of　spatial　information,　which　appear　to　provide　a　basis　for　cognitive　map:　a　representation　of　environment.　Most　prior　biomimetic　map　building　algorithms,　such　as　RatSLAM　algorithm　or　traditional　SLAM　methods,　have　little　biological　fidelity　to　the　hippocampal　formation.　In　this　paper,　a　neural　network　model　based　on　the　behavioral　and　neurophysiological　mechanisms　of　the　spatial　cells　is　constructed,　and　is　applied　to　building　the　accurate　cognitive　map　of　real　environments.　The　proposed　algorithm　has　a　uniform　calculation　method　for　spatial　cells　based　on　continuous　attractor　network　dynamics　to　integrate　self-motion　cues,　which　can　reproduce　grid　cells　firing　responses　and　place　cells　firing　fields　via　feedforward　inputs　from　band　cells.　RGB-D　images　serve　as　visual　cues　for　loop　closure　detection　and　correcting　the　accumulative　errors　intrinsically　associated　with　the　path　integration　mechanism,　which　contributes　to　building　spatial　cognitive　maps　of　indoor　environments　on　a　mobile　robot.　A　cognitive　map　is　a　fine-grained　topological-metric　map.　A　node　in　the　cognitive　map　is　constructed　by　associating　the　major　peak　of　place　cell　population　activities　with　corresponding　visual　cues　and　the　transition　stores　the　change　in　positions.　Simulation　experiments　and　physical　experiments　with　a　mobile　robot　have　verified　the　effectiveness　of　the　algorithm.　The　proposed　algorithm　may　provide　a　foundation　for　robotic　navigation.　Copyright　©　2018　Acta　Automatica　Sinica.　All　rights　reserved.