Training alone together: developing a virtual opponent for home-based kickboxing training

McCrossan, James and Stawarz, Katarzyna ORCID: https://orcid.org/0000-0001-9021-0615 2026. Training alone together: developing a virtual opponent for home-based kickboxing training. Presented at: BCS-HCI: British Computer Society Conference on Human-Computer Interaction, London, UK, 1-3 November 2026. Proceedings of BCS HCI 2026. BCS Learning and Development Ltd, Item availability restricted.

PDF - Accepted Post-Print Version Restricted to Repository staff only until 1 November 2026 due to copyright restrictions. Download (657kB)

Abstract

Solo training leaves martial artists without a resource it cannot replace: a responsive sparring partner - an absence the COVID-19 pandemic made acute when it closed gyms and dojos worldwide. Replicating that partner digitally requires real-time recognition of a practitioner’s movement, which is still an open challenge for SportsHCI. Virtual Reality (VR) is one route: consumer headsets can place a digital opponent in front of the practitioner, though few actually adapt to how the user is moving. To address this gap, drawing from our experiences as martial arts practitioners, we present a proof-of-concept VR sparring system for kickboxing that attempts to close a feedback loop of recognition, prediction and response for a solo practitioner. Webcam pose estimation feeds a machine learning (ML) classifier, whose output drives an ML sequence predictor that selects the response of an avatar opponent in Unity, a 3D game engine. The classifier reached 87.22% test accuracy across 18 movement classes but reliably recognised only three in deployment. Assessed through autobiographical design, the system is presented as an early exploration from which we draw two provisional design principles for responsive digital sparring partners: calibrated unpredictability and a readable opponent.

Item Type:	Conference or Workshop Item - published (Poster)
Status:	In Press
Schools:	Schools > Computational & Mathematical Sciences Schools > Computer Science & Informatics
Publisher:	BCS Learning and Development Ltd
Date of First Compliant Deposit:	14 September 2026
Date of Acceptance:	21 August 2026
Last Modified:	15 Sep 2026 14:00
URI:	https://orca.cardiff.ac.uk/id/eprint/189557

Actions (repository staff only)

Edit Item