Developing an Interactive Virtual Reality Dance Partner

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Date

2024-08-19

Department

Major/Subject

Computer Science

Mcode

SCI3042

Degree programme

Master’s Programme in Computer, Communication and Information Sciences

Language

en

Pages

43

Series

Abstract

Virtual Reality (VR) offers enhanced immersion and potential for complex interactions with the virtual space and characters, allowing to utilise the whole body of the user as an input data source. Additionally, this technology has been seeing wider adoption by users over the years. As such, this makes VR a perfect medium for exploring casual embodied dancing with interactive virtual agents. However, designing experiences for VR also reveals a landscape of technical problems to be solved, especially in character animation. This thesis focuses on real-time embodied VR interaction in the domain of dance with the goal of developing an interactive VR dance partner. The key technical problem to solve is how to animate this character in order to produce high-quality dance movements and a positive user experience, all in real-time. In VR, the increased immersion may result in users being able to notice animation errors more easily, thus requiring more sophisticated animation approaches. At the same time, computing power constraints for real-time use demand of the system to be computationally light-weight and performant, as the VR devices holding the largest market share among users are standalone devices. While novel Generative Artificial Intelligence (AI) solutions are showing promise in character synthesis, the real-time requirement constraint means that character animation in this context is still largely an open issue. Here, we present our work on implementing a virtual dance partner in a VR application prototype utilising an animation approach called Motion Matching - an industry standard approach that produces new animation sequences at run-time from available animation data. We conclude that Motion Matching is a viable and general approach for animating interactive dancing agents in VR. Additionally, we find that Motion Matching in this context also allows for context-specific optimizations, such as structuring the data by beat type while simultaneously encoding footwork, and synchronising animation queries to the beat of the music.

Description

Supervisor

Hämäläinen, Perttu

Thesis advisor

Hämäläinen, Perttu

Keywords

virtual reality, character animation, motion matching, dancing

Other note

Citation