One-Shot Learning for Robotic Manipulators: Rapid Replication of Human Activities from a Single Demonstration

This paper presents a One-Shot Learning framework able to process a RGB-D video of a human task demonstration and to perform it on a robot manipulator. Learning from a single human demonstration is one of the most interesting challenges in robotics. The aim is to allow a robot to reproduce operator's activities after observing how they are performed just once. Although the work presented in this paper focuses on specific manipulation tasks, the proposed method can be extended to multi-stage operations carried out in different fields, both domestic and industrial. In the proposed approach, the demonstration is first segmented into primitives, which are then mapped into robot actions to be executed by a manipulator. This work also aims to ensure that the learning process is carried out rapidly. The paper provides an overview of the overall framework and illustrates the system at work in a use case.