Personalized Telerobotics by Fast Machine Learning of Body-Machine Interfaces

Abstract

Human-Robot Interfaces (HRIs) can be hard to master for inexperienced users, making the teleoperation of mobile robots a difficult task. The development of Body-Machine Interfaces (BoMIs) represents a promising approach to making a user more proficient, by exploiting the natural control they can exert on their own body motion. Since human motion presents individual traits due to several factors, including physical condition, age, and experience, generic BoMIs still require a significant learning time and effort to reach adequate ability in teleoperation. In this work, we present a novel approach which provides a Body-Machine Interface tailored on the specific user. Our method autonomously learns from the user their preferred strategy to control the robot, and provide a personalized body-machine mapping. We show that the proposed method can significantly reduce the duration of the training phase in teleoperation, thus allowing faster skill acquisition. We validated our approach by performing both simulation and real-world experiments with human subjects. The first involved the teleoperation of a fixed-wing simulated drone, while the second consisted in controlling a real quadrotor. We used our framework to extrapolate common and peculiar features of movements among individuals. Observing reoccurring strategies, we provide insights on how humans would naturally interface with a distal machine.