A　fuzzy　operant　conditioning　probabilistic　automaton(OCPA)　bionic　autonomous　learning　system　is　constructed　based　on　Skinner　operant　conditioning　theory　and　combined　with　the　probabilistic　automaton　and　fuzzy　inference　for　realizing　a　two-wheeled　robot　self-balancing　control.　The　learning　system　is　a　stochastic　mapping　from　state　sets　to　operant　action　sets.　The　optimal　action　for　controlling　the　system　is　stochastically　learned　from　the　operant　action　set　by　adopting　operant　conditioning　learning　algorithm;　in　the　same　time　the　orientation　value　information　of　the　learned　operant　action　is　used　to　adjust　the　operant　conditioning　learning　algorithm.　In　addition,　the　action　entropy　is　added　to　verify　the　self-learning　and　self-organizing　ability　of　the　learning　system.　In　the　simulation,　a　two-wheeled　robot　self-balancing　control　is　realized,　demonstrating　the　feasibility　of　the　fuzzy　OCPA　learning　system.