Abstract
To realize the skill migration and generalization of robotic arm in industrial production. We extracts 7 basic movement primitives from industrial tasks. Parameterize each primitive and build a library of movement primitives, and give the connection mode between primitives. Then, the robotic arm Behavior Tree (BT) is constructed according to the execution logic of the task. When faced with a new task, use Dynamic Movement Prmitives (DMP) to generalize the movement primitives according to the target pose. When faced with an unknown environment, control the action by selecting a specific BT. Finally, the effectiveness of the framework is verified through experiments.