Investigation into Recording, Replay and Simulation of Interactions in Virtual Reality
Michael Siebenmann, Mathieu Lutfallah, Dominic Jetter, Christian Hirt, Andreas M. Kunz · 2024
Previous work has shown various authoring toolkits relying on recording, but quantitative comparisons between the 3D recording approaches remain unexplored. In this study, we introduce an authoring toolkit that allows both experts and trainees to record their actions in a virtual environment, streamlining the creation of training procedures and the evaluation of trainee actions. The toolkit was developed using two distinct methods: a state-based and an input-based approach. Within a virtual testing environment, we compared these methods across a range of interactions, focusing on three performance metrics: memory footprint, performance overhead, and replay accuracy. Contrary to initial predictions, the state-based method, after optimization, consumed less memory than the input-based approach. Both methods maintained low performance overheads during recording and replaying phases. Notably, the state-based approach achieved superior replay accuracy. In contrast, the input-based method displayed varying degrees of replay inaccuracies, partially attributable to the non-deterministic physics engine of the Unity development platform.