Learning robot behaviour and skills based on human demonstration and advice: The machine learning paradigm

Service robots require easy programming methods al-lowing the unexperienced human user to easily inte-grate motion and perception skills or complex prob-lem solving strategies. To achieve this goal, robots should learn from operators how and what to do con-sidering hard- and software constraints. Various ap-proaches modelling the man-machine skill transfer have been proposed. Systems following the Program-ming by Demonstration (PbD) paradigm that were de-veloped within the last decade are getting closer to this goal. However, most of these systems lack the possi-bility for the user to supervise and in uence the pro-cess of program generation after the initial demonstra-tion was performed. In this paper a principle learning methodology is discussed, which allows to transfer hu-man skills and to supervise the learning process includ-ing subsymbolic and symbolic task knowledge. Here, several existing approaches will be discussed and com-pared to each other. Moreover, a system approach is presented, integrating the overall process of skill trans-fer from a human to a robotic manipulation system. One major goal is to modify information gained by the demonstration in that way that dierent target sys-tems are supported. The resulting PbD-system yields towards a hybrid learning approach in robotics to sup-port natural programming based on human demonstra-tions and user advice.