2019/10/14 by Judith Bütepage, Bütepage, Judith, Ali Ghadirzadeh +7 · 2 citations
Computer Science · Engineering · Psychology · #FOS: Computer and information sciences #Human Motion and Animation #Human Pose and Action Recognition #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Robotics (cs.RO) #Social Robot Interaction and HRI
paper · pdf · doi:10.48550/arxiv.1910.06031
openalex publication_date 2019/10/14 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
To coordinate actions with an interaction partner requires a constant\nexchange of sensorimotor signals. Humans acquire these skills in infancy and\nearly childhood mostly by imitation learning and active engagement with a\nskilled partner. They require the ability to predict and adapt to one's partner\nduring an interaction. In this work we want to explore these ideas in a\nhuman-robot interaction setting in which a robot is required to learn\ninteractive tasks from a combination of observational and kinesthetic learning.\nTo this end, we propose a deep learning framework consisting of a number of\ncomponents for (1) human and robot motion embedding, (2) motion prediction of\nthe human partner and (3) generation of robot joint trajectories matching the\nhuman motion. To test these ideas, we collect human-human interaction data and\nhuman-robot interaction data of four interactive tasks "hand-shake",\n"hand-wave", "parachute fist-bump" and "rocket fist-bump". We demonstrate\nexperimentally the importance of predictive and adaptive components as well as\nlow-level abstractions to successfully learn to imitate human behavior in\ninteractive social tasks.\n