Logo image
Back
Predicting Ground Reaction Forces by LSTM Neural Networks with Multi-modal Data
Conference proceeding

Predicting Ground Reaction Forces by LSTM Neural Networks with Multi-modal Data

Xinyi Fu, Zhaoyuan Wan, Zhizhang Li, Suiyuan Wang, Te Zhang, Hui Zhang, Longbin Zhang, Xueyu Zhu and Ruoli Wang
Youth Academic Annual Conference of Chinese Association of Automation (Online), pp.2041-2046
05/08/2026
DOI: 10.1109/YAC71005.2026.11615719

View Online

Abstract

Accurate estimation of three-dimensional ground reaction force in non-laboratory settings is pivotal for biomechanical analysis and rehabilitation monitoring. While wearable sensors provide a portable solution, single-modality approaches relying solely on Inertial Measurement Units (IMUs) or pressure insoles often encounter information deficits, leading to limited estimation accuracy in complex dynamic tasks. To address this limitation, this study proposes a multi-modal fusion framework based on a bidirectional long short-term memory network. The framework integrates whole-body Inverse Kinematics (IK) data derived from IMUs with plantar pressure distribution data from smart insoles. We validated the model under four distinct evaluation protocols on a dataset comprising nine healthy subjects, covering a diverse range of daily functional activities, including walking, squatting, sit-to-stand, transitions, and vertical jumps. Experimental results demonstrated that the multi-modal fusion method consistently outperforms single-modal baselines. Furthermore, the model exhibited adaptability across diverse motor tasks, maintaining consistent performance even during complex transient activities involving rapid center-of-mass modulation. These results suggest that combining IK and plantar pressure data provides complementary biomechanical information, thereby mitigating the limitations associated with single-modality estimation and improving the generalizability of gait analysis in free-living environments.
 Machine Learning Aluminum Bidirectional long short term memory Estimation Force ground reaction forces Grounding Image sensors kinetics Modeling Printing Protocols sensing insole Wearable devices wearable sensors

Details

Metrics

1 Record Views
Logo image