One-shot robot programming by demonstration by adapting motion segments

Abstract

In order to provide robots for a wide range of users, they must be easily programmable. This is addressed by various approaches of programming a robot by demonstration, where the user guides a robot through the task. We propose an approach that adapts a once demonstrated trajectory to a new situation. The system extracts the relevant reference frames using a weighting function. Then, the demonstrated trajectory is segmented with respect to these frames and different frame-dependent strategies for reproduction are applied. The approach is capable of adapting simple point-to-point paths, more complex tasks like pick-and-place operations, and manipulations where objects are used as tools. Additionally, the system extracts the required resources for a task to integrate the approach into a behavior-based system, which manages new programmed behaviors and selects a suitable behavior in a new situations. The approach is computationally efficient and comprehensible for the user. We evaluated the approach qualitatively and quantitatively using cross validation.