Aiming　at　the　movement　balance　control　problem　of　the　two-wheeled　self-balancing　mobile　robot,　a　bionic　self-learning　algorithm　consisting　of　BP　(backpropagation)　neural　network　and　eligibility　traces　based　on　Skinner＇s　operant　conditioning　theory　is　put　forward　as　a　learning　mechanism　of　the　two-wheeled　robot.　The　algorithm　utilizes　the　characters　of　eligibility　traces　in　resolving　delay　effect,　increasing　learning　speed,　and　improving　reliability　and　ability,　so　that　the　complex　learning　algorithm　consisting　of　BP　neural　network　and　eligibility　traces　can　predict　the　behavior　evaluation　function　that　the　robot　would　obtain,　and　choose　the　optimum　action　corresponding　to　the　biggest　evaluation　value　according　to　the　probability　tendency　mechanism　by　a　certain　probability.　Thereby　the　two-wheeled　robot　can　obtain　the　self-learning　skills　like　a　human　or　animal　by　interacting　with,　studying　and　training　the　unknown　environment,　and　realize　the　movement　balance　control　of　the　two-wheeled　robot.　Finally,　two　simulation　experiments　are　done　and　compared　using　the　BP　algorithm　and　the　complex　learning　algorithm　consisting　of　BP　neural　network　and　eligibility　traces　based　on　Skinner＇s　operant　conditioning　theory.　The　simulation　results　show　that　the　learning　mechanism　of　the　complex　learning　algorithm　consisting　of　BP　neural　network　and　eligibility　traces　based　on　Skinner＇s　operant　conditioning　theory　makes　the　robot　obtain　the　better　dynamic　performance　and　the　quicker　learning　speed,　and　reflect　stronger　self-learning　skills　and　balance　control　abilities.