%0 Journal Article
%T Data-Driven Estimation of Rowing Forces and Power via Long Short-Term Memory Neural Networks
%A Lucas Meyer
%A Anna Schmid
%A Stefan Braun
%J Bulletin of Pioneering Researches of Medical and Clinical Science
%@ 3006-2659
%D 2024
%V 4
%N 1
%R 10.51847/GJR2UDTUZl
%P 186-200
%X Assessing rowing output, such as by examining force and power delivery curves from an ergometer or a boat, is a top priority for coaches and athletes alike. The gold-standard approaches currently available for rowing output evaluation require purpose-built sensorized hardware, with PowerLine and BioRow being the most widely adopted options. This workflow is both financially demanding and labor-intensive, thereby reducing the frequency with which coaches can oversee rowers. In this work, we devised a simpler-to-mount, lower-cost technique for inferring rowers’ forces and powers using only cable-position transducers for indoor rowing and inertial measurement units (IMUs) and GPS for on-water sculling. Recordings from 12 and 11 rowers, on an ergometer and in a boat, respectively, were used to learn the parameters of a long short-term memory (LSTM) network. The LSTM proved capable of recovering gate forces and power, with a total mean absolute error remaining below 5%. The recovered force and power profiles uncovered technical differences between individuals, attaining 93% accuracy. Undertaking leave-one-out cross-validation resulted in a substantial increase in error, supporting the conclusion that a broader pool of participants is necessary to yield a model that generalizes to unseen rowers.
%U https://bprmcs.com/article/data-driven-estimation-of-rowing-forces-and-power-via-long-short-term-memory-neural-networks-mxp8v3ttc3kz79l