Data-driven modelling of low-dimensional dynamical structures underlying complex full-body human movement

A new study uses Neural Ordinary Differential Equations (NODEs) to model complex, full-body human movements like a baseball pitch as a low-dimensional dynamical system. The model could accurately predict the time evolution of a pitching motion (R² > 0.45). Crucially, it explained approximately 50% of the variance in the latter half of the motion using only the initial ~8% of the sequence, showing movement evolves from initial conditions in a latent space.