Estimation of Ground Reaction Forces in Rugby Scrummaging Using Instrumented Insoles and Machine Learning

Scrummaging is a critical and intense phase of rugby union, involving considerable forces and presenting the highest risk of injury during a rugby game. Despite its importance as an indicator of game success, the forces generated by individual players within the scrummaging have yet to be investigated. Studies have mainly focused on individual and collective pushing forces against scrummaging machines, with little attention to the ground reaction forces. However, these ground reaction forces are essential for dynamic biomechanical analyses, particularly in sports performance and reduction of injury risks. While instrumented insoles have been validated for gait and movements involving primarily vertical forces, their application in rugby scrummaging remains unexplored. The objective of this paper was to develop a method to estimate the three components of ground reaction forces, using a combination of instrumented insoles and machine learning. A Long-Short-Term-Memory (LSTM) neural network was trained to estimate ground reaction forces from data collected by instrumented insoles and inertial measurement units. Thirteen participants were evaluated to validate the method, performing thrusts against an individual scrum machine with feet on force plates. The average root mean square errors over all subjects between the estimated forces and the reference force plate data were 58.4±12.0N on the Antero-Posterior axis, 21.2±2.0N on the Medio-Lateral axis, and 61.1±9.2N on the vertical axis. These findings demonstrate the potential of using instrumented insoles to accurately estimate the ground reaction forces, particularly the thrust force, of rugby players during scrummaging.