Autonomous Robots https://doi.org/10.1007/s10514-018-9699-4 One-shot learning of human–robot handovers with triadic interaction meshes David Vogt 1 · Simon Stepputtis 2 · Bernhard Jung 1 · Heni Ben Amor 2 Received: 31 December 2016 / Accepted: 13 January 2018 © Springer Science+Business Media, LLC, part of Springer Nature 2018 Abstract We propose an imitation learning methodology that allows robots to seamlessly retrieve and pass objects to and from human users. Instead of hand-coding interaction parameters, we extract relevant information such as joint correlations and spatial relationships from a single task demonstration of two humans. At the center of our approach is an interaction model that enables a robot to generalize an observed demonstration spatially and temporally to new situations. To this end, we propose a data-driven method for generating interaction meshes that link both interaction partners to the manipulated object. The feasibility of the approach is evaluated in a within user study which shows that human–human task demonstration can lead to more natural and intuitive interactions with the robot. Keywords Human–human demonstration · Human–robot interaction · Handover · Interaction mesh 1 Introduction Handing over an object to another person is arguably one of the most essential physical interaction skills. Independently of whether we are at home, in the workplace, at a restaurant, or at the hospital, we are often faced with situations in which we either receive an object, or handover an object to another person. Hence, for robots to be reliably used as assistants to humans, they have to be able to engage in similar inter- actions and deal with the large variability inherent to such This is one of the several papers published in Autonomous Robots comprising the Special Issue on Learning for Human–Robot Collaboration. B David Vogt contact@david-vogt.com Simon Stepputtis sstepput@asu.edu Bernhard Jung bernhard.jung@informatik.tu-freiberg.de Heni Ben Amor hbenamor@asu.edu 1 Faculty of Mathematics and Informatics, Technical University Bergakademie Freiberg, 09599 Freiberg, Germany 2 School of Computing, Informatics, Decision Systems Engineering, Arizona State University, 699 S Mill Ave, Tempe, AZ 85281, USA handover tasks. Hand-overs are joint tasks in which the giver and receiver coordinate their movements in order to ensure the successful transition of the object from one to the other (see Fig. 1). This requires the interaction partners to react and adapt to each others’ movement, timing, style, and posture. With the advent of collaborative robots, research on human–robot handovers has found increased interest in the robotics community. Various strategies for specifying and learning such behavior have been put forward, such as in Duvallet et al. (2016); Ewerton et al. (2015). While these approaches have produced important insights, they mostly model human–robot handover as a dyadic interaction process—the process parameters are solely influenced by the two interaction partners and not the handled object. However, especially in situations in which an object is handed from a human to a robot, it is important to incorporate the object as an additional element in the interaction process. In addition, the majority of approaches focuses on the spatial relationship of the end-effectors during the task. Only the position of the human hand is used to identify the robot’s response. In this paper, we propose a methodology for learning triadic interaction meshes from observed human–human demonstrations. In particular, we focus on scenarios in which the robot receives an object from a human partner. Given a single demonstration, we can extract information about the synchrony in movement between different body parts of the two interactants, spatial relationships between inter- 123