Qantifying Coordination in Human Dyads via a Measure of Verticality Roshni Kaushik Mechanical Science and Engineering Champaign-Urbana, Illinois rkaushi2@illinois.edu Ilya Vidrin Movement Lab Cambridge, Massachusetts ilya_vidrin@mail.harvard.edu Amy LaViers Mechanical Science and Engineering Champaign-Urbana, Illinois alaviers@illinois.edu ABSTRACT Working towards the goal of understanding complex, interactive movement in human dyads, this paper presents a model for analyz- ing motion capture data of human pairs and proposes measures that correlate with features of the coordination in the movement. Based on deep inquiry of what it means to partner in a motion task, a measure that characterizes the changing verticality of each agent is developed. In parallel a naïve human motion expert provides a qual- itative description of the features and quality of coordination within a dyad. Analysis on the verticality measure, the cross-correlation of verticality signals, and deviation of those verticality signals from the trend over time, provides quantitative insight that corroborates the naïve expert’s analysis. Specifcally, the paper shows that, for four samples of dyadic behavior, these measures provide informa- tion about 1) whether two agents were involved in the same dyadic interaction and 2) the level of "resistance" found in these interac- tions. Future work will test this model over a larger dataset and develop human-robot coordination schemes based on this model. CCS CONCEPTS · Human-centered computing → Empirical studies in inter- action design; Empirical studies in interaction design; · Ap- plied computing → Performing arts; KEYWORDS motion-capture, robotics, partner, interaction, coordination, dyad ACM Reference Format: Roshni Kaushik, Ilya Vidrin, and Amy LaViers. 2018. Quantifying Coordina- tion in Human Dyads via a Measure of Verticality. In MOCO: 5th International Conference on Movement and Computing, June 28–30, 2018, Genoa, Italy, Jen- nifer B. Sartor, Theo D’Hondt, and Wolfgang De Meuter (Eds.). ACM, New York, NY, USA, Article 4, 8 pages. https://doi.org/10.1145/3212721.3212805 1 INTRODUCTION Human movement is a complex physical phenomenon, full of the richness of contexts, interactions, and variations. In particular, the intricacies of interactive movement raise many research questions, including the manner of nonverbal communication between a pair Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. MOCO, June 28–30, 2018, Genoa, Italy © 2018 Association for Computing Machinery. ACM ISBN 978-1-4503-6504-8/18/06. . . $15.00 https://doi.org/10.1145/3212721.3212805 or dyad performing a task together. In something as simple as moving a table across a room, two individuals communicate through the movement of their bodies in addition to the forces applied on the table and the foor. In partner dance, this communication channel is even more nuanced. When dissecting interactive human movement, we seek to identify properties describing and characterizing these interactions. A number of studies have been conducted on dyadic interactions. For example, analyzing the making and breaking of symmetry of the head (mirror symmetry) during conversations showed to be a meaningful element of communication when modeled with a dynamical system [5, 7, 8]. Clinicians found that understanding micro-movements using kinematic recordings could allow them to classify dyadic interactions of people with social difculties more quantitatively [29]. Additionally, movement as an important design aspect in human-computer interaction prompted a course on embodied interaction, formalizing the applications for many movement aspects [11]. Categorizing the large variety of movement can draw analogies from studies that look for parameterizations of other large datasets. The search for a parameterization of images using thermodynamic principles such as energy and entropy drew many parallels between the physical intuition of thermodynamics and properties of the im- age, revealing measures that refected natural versus urban images [26]. An interactive online dance work allowed researchers to better understand the interactions between the audience and the work and develops kinesthetic empathy as a parameter in movement representations [10]. Machine learning and neural networks can be used to abstract away the complexities of interaction by training models with ex- amples. Gaussian Mixture Models (GMM) of Interaction Primitives model nonlinear correlations between diferent movers [12, 18]. Task-parameterized dynamical systems combined with learning allowed a robot to learn a collaborative task after observing a pair of humans performing the same task [24]. A GMM trained with examples of two humans interacting recognized new actions and generated responses of a virtual character [13]. Learning from demonstrations, a virtual dancer developed an internal model of a human dancer’s movements using Artifcial Neural Networks (ANN) and Hidden Markov Models (HMM) and reacted to some movements from a human dancer [19]. Haptic feedback, a way to measure the forces a user exerts on an interface, is another tool used to understand interactive mo- tion. A dancing robot adjusted the length of its stride based on haptic feedback from the physical connection between robot and human [27], and male and female partner dancing behavior was synthesized based on haptic interactions and stride length [14]. A