This repo contains the code of the Expectation Maximization-based clustering algorithm adapted slightly from [1]. It has been used to cluster the observation data obtained from human-human interactions of professional squash coaches and stroke physiotherapists. These observations produced action sequences (e.g. pre-instruction, praise, concurrent instruction (positive), positive modelling, questioning, post instruction (positive)) from which transisiton matrices were produced. These transition matrices are used as the data points as input to this clustering program.
The full process is explained further in the paper's below.
[1] Stefanos Nikolaidis, Ramya Ramakrishnan, Keren Gu, and Julie Shah. [n. d.]. Efficient Model Learning from Joint-Action Demonstrations for Human-Robot Collaborative Tasks. In Proceedings of the Tenth Annual ACM/IEEE International Conference on Human-Robot Interaction - HRI ’15 (2015-03). ACM, 189–196. https://dl.acm.org/doi/10.1145/2696454.2696455
[2] Ross, Martin K., Frank Broz, and Lynne Baillie. ‘Observing and Clustering Coaching Behaviours to Inform the Design of a Personalised Robotic Coach’. In Proceedings of the 23rd International Conference on Human-Computer Interaction with Mobile Devices and Services. Virtual (originally Toulouse, France): ACM, 2021. https://doi.org/10.1145/3447526.3472043.