Design of Expressive Motion Primitives for Safe Human-Robot Partner Dancing

Abstract

As part of the Safe Robotics Lab’s Human-Robot Partner Dancing Project, this thesis describes the design of a reference trajectory generator and cost functions, to be used in the creation of a library of partner dance motion primitives for an Agility Robotics Digit robotic platform. It details the design of a dance move description system that can compactly encode a large variety of partner dance steps, and was used to generate reference trajectories for the feet of the robot to perform these steps. The description system was intentionally constructed so that multiple forms of expression can be easily added to the step trajectories, in order to give the robot the ability to match its human partner’s style and pace, and also effectively communicate its plans during the dance. The reference trajectories were used to parametrize a family of cost functions for trajectory optimization based on an Asymmetric Spring Loaded Inverted Pendulum (ASLIP) model of robotic bipedal locomotion in three-dimensional space, which was used to simplify the complex dynamics of Digit’s locomotion. The cost functions were designed so that they can be easily adapted for Digit, and then utilized by a Hybrid Iterative Linear-Quadratic Regulator (ILQR) trajectory optimization algorithm in future steps of the project.