Global　path　planning　in　more　general　scene　under　the　constraint　condition　has　always　been　the　difficulty　in　the　field　of　path　planning.　Rapidly-exploration　Random　Tree　and　its　derived　algorithm　(RRTs)　is　a　popular　planning　algorithm　of　path　planning　in　recent　years,　which　is　especially　suitable　for　constrained　conditions.　A　feasible　path　should　consider　the　model　and　constraints　of　its　execution　object.　To　this　end,　we　propose　S-BRRT*:　First,　a　bidirectional　spanning　tree　is　introduced　into　the　basic　RRT*.　Second,　nonholonomic　constraints　are　considered　into　the　program,　including　internal　motion　constraints　and　external　collision　constraints.　Then,　a　new　pruning　strategy　is　put　forward　to　conduct　intelligent　pruning　of　BRRT*　＇s　zigzag　path,　including　pruning　under　collision　constraint　and　pruning　under　path　constraint,　so　as　to　select　optimal　path　nodes.　Finally,　the　smooth　strategy　is　proposed,　using　cubic　Bezier　curve　to　fitting　the　previous　pruning　path　for　the　first　time.　Then,　we　fitting　for　the　second　time　under　collision　optimization　to　ensure　the　final　feasibly　smooth　path.　Our　S-BRRT　*　algorithm　can　realize　the　global　path　planning　in　a　variety　of　scenarios.　It　is　verified　by　experiments　in　sparse　common　scene　as　well　as　the　scenarios　with　dense　narrow　path.　We　can　conclude　that　the　trajectory　maintains　robust　and　conforms　to　the　nonholonomic　constraints,　meanwhile,　we　obtain　improvement　in　length　and　feasibility.