Modern industrial applications require robots to be able to operate in\nunpredictable environments, and programs to be created with a minimal effort,\nas there may be frequent changes to the task. In this paper, we show that\ngenetic programming can be effectively used to learn the structure of a\nbehavior tree (BT) to solve a robotic task in an unpredictable environment.\nMoreover, we propose to use a simple simulator for the learning and demonstrate\nthat the learned BTs can solve the same task in a realistic simulator, reaching\nconvergence without the need for task specific heuristics. The learned solution\nis tolerant to faults, making our method appealing for real robotic\napplications.\n