AI-based human whole-body posture-prediction in continuous load reaching/leaving activities
Reza Ahmadi1, Mahdi Mohseni1, Navid Arjmand1
1Department of Mechanical Engineering, Sharif University of Technology, Tehran, Iran.
Journal of Biomechanics
|April 13, 2025
Summary
This study developed artificial neural networks (ANNs) to predict worker body posture during load handling, improving ergonomic assessments. The ANNs offer continuous, phase-specific posture prediction for occupational risk management.
Area of Science:
- Occupational health and safety
- Biomechanics
- Artificial intelligence in ergonomics
Background:
- Assessing worker body posture is crucial for managing musculoskeletal injury risk.
- Traditional posture measurement methods are impractical for real-world occupational settings.
- Existing models lack continuous, phase-specific posture prediction for dynamic tasks.
Purpose of the Study:
- To develop artificial neural networks (ANNs) for predicting 3D continuous full-body posture.
- To focus on load-reaching and load-leaving phases of lifting/lowering activities.
- To complement previous ANNs for the load-moving phase.
Main Methods:
- Utilized a whole-body motion dataset from 20 healthy novice subjects.
- Developed four ANNs using task- and subject-specific parameters as inputs.
- Estimated continuous body coordinates and segment/joint angles.
Main Results:
- Achieved root-mean-square errors <3 cm and <10° for load-reaching.
- Achieved root-mean-square errors <4 cm and <15° for load-leaving.
- Identified higher errors on the left body side and in the latter activity halves.
Conclusions:
- The developed ANNs enable continuous, phase-specific posture prediction, enhancing ergonomic applications.
- This approach offers a step towards accessible posture prediction tools in occupational settings.
- Further research is needed across diverse demographics for broader applicability.
