Evaluating the Performance of Deep Learning Models in Whole-body Dynamic 3D Posture Prediction During Load-reaching Activities
Researchers have developed deep learning models, specifically BLSTM and transformer architectures, to predict human body posture during dynamic load-reaching activities. The models utilize hand-load position, lifting techniques, and initial body posture data to forecast subsequent movements. A novel cost function was introduced to enforce constant body segment lengths, improving prediction accuracy by up to 21%. The transformer model demonstrated superior performance, achieving a root-mean-square error of 41.4 mm and outperforming the BLSTM model by approximately 58% in long-term prediction. AI
IMPACT Introduces improved AI methods for predicting human motion dynamics in manual material handling.