Learning　is　the　main　aim　of　robotics.　In　this　paper　we　present　a　new　stochastic　learning　automaton　called　a　Skinner　automaton　as　a　psychological　model　for　formalizing　the　theory　of　operant　conditioning.　We　identify　animal　operant　learning　with　a　thermodynamic　process,　and　derive　a　so-called　Skinner　algorithm　from　Monte　Carlo　method　and　Metropolis　algorithm　and　simulated　annealing.　The　Skinner　automaton　is　implemented　on　a　two-wheeled　robot　with　a　flexible　lumbar　in　a　simulation　experiment　and　it　learns　to　keep　balance　successfully.