Robot Imitation of Polishing Motions by Observing Humans: From Human Non-Verbal Cues to Stable Limit Cycles

Humans and robots are progressing towards a shareable and co-existing workplace environment. To progress in that direction robots must understand how humans operate as well as their intention at each situation, while at the same time, humans must be aware of the robot's capabilities, and in some similarities, the intention of movement. We propose a time-invariant dynamical system with a limit cycle attractor as the model for all circular/ellipse type polishing motions. This approach was tested on a human dataset of polishing trajectories and in a human-robot scenario resulting in fast and accurate reproduction of the observed polishing motions. The human-robot scenario is accomplished with an end-to-end system that allows for real-time human demonstration of a polishing strategy of a table, an optimization solver that finds the best match for the dynamical system, and the optimal limit cycle is executed by the robot to polish the same table. The code and dataset are available at: https://github.com/NunoDuarte/RobotPolishingDS.