The　rise　of　shared　bicycles　has　increased　the　demand　for　group　riding,　integrating　bicycles　into　social　groups.　Additionally,　retrograde　riding,　where　cyclists　travel　against　the　designated　direction,　is　a　common　behavior　observed　in　bicycle　flows.　The　interaction　and　self-organization　phenomenon　of　group　and　retrograde　behaviors　are　complex,　significantly　impacting　traffic　efficiency.　This　paper　develops　a　two-dimensional　Extended　Moore　Neighborhood　and　constructs　state-updating　rules　for　regular　riding,　group　riding　and　retrograde　riding.　Each　rule　comprises　a　psychological　decision　layer　and　a　physical　execution　layer,　forming　a　cellular　automaton　model　for　group　and　retrograde　bicycles.　Field　experiments　are　conducted　to　calibrate　the　model　parameters　and　verify　the　behavioral　characteristics.　Finally,　we　execute　numerical　simulations　at　a　signalized　intersection　to　explore　the　coupling　effects　of　group　and　retrograde　behaviors　on　selforganization　within　the　bicycle　flow　and　the　traffic　capacity.　The　results　indicate　that　group　behavior　increases　queue　length　while　reducing　start　wave　speed　and　expansion　degree.　Retrograde　behavior　intensifies　the　negative　effects　on　bicycle　flow.　These　findings　provide　insights　for　managing　both　forward　and　retrograde　bicycle　flows.