Reproduce “UnderPressure: Deep Learning for Foot Contact Detection, Ground Reaction Force Estimation and Footskate Cleanup” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lucas Mourot, Ludovic Hoyet, François Le Clerc, Pierre Hellier
Researchers from Inria, Univ Rennes, CNRS, and InterDigital developed a deep learning approach for robust foot contact detection and footskate cleanup, utilizing estimated Vertical Ground Reaction Forces (vGRF) from motion capture data. Their method, which achieved an F1 score of 0.947 for contact detection and an RMSE of 10.9% for vGRF estimation, incorporates a novel vGRF invariance constraint to enhance the physical plausibility of cleaned motions.
Key insights
- Modeling continuous physical quantities like vGRFs fosters a deeper, more dynamically informed understanding of foot-ground interaction in deep neural networks, leading to more robust contact detection compared to direct binary classification.
- Robustness to motion perturbations is significantly improved through comprehensive data augmentation strategies, including stochastic transformations and randomization of skeleton morphologies during training.
- Enforcing vGRF invariance as a constraint during inverse kinematics optimization is crucial for guiding footskate cleanup, ensuring that corrected motions remain physically plausible and consistent with the original foot-ground dynamics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.