Tactile Gloves Predict Load Weight During Lifting With Deep Neural Networks.

Area of Science:

• Occupational Safety
• Biomechanics
• Machine Learning

Background:

• Overexertion during lifting is a major cause of workplace injuries.
• Accurate load weight estimation is crucial for assessing lifting task risks.
• Current methods for estimating load weight are often indirect or require intrusive sensors.

Purpose of the Study:

• To propose tactile gloves as a novel method for predicting load weight during lifting.
• To evaluate the performance of deep neural networks in estimating load weight from tactile data.
• To analyze hand force exertion patterns during lifting using explainability techniques.

Main Methods:

• Tactile gloves were used to collect hand pressure data during lifting tasks.
• Collected data were formulated into a 2-D matrix capturing spatial and temporal information.
• A ResNet 18 deep neural network regression model was employed for load weight prediction.
• Shapley additive explanations (SHAPs) were used to interpret model decisions and identify key features.

Main Results:

• The ResNet 18 model achieved a predicted R-squared of 0.821 and a mean absolute error of 1.579 kg.
• SHAPs analysis revealed that the right hand, fingers, and the middle lifting phase were most important for weight prediction.
• The study demonstrated the feasibility of using tactile sensing for load weight estimation.

Conclusions:

• Tactile gloves offer a viable, non-intrusive method for predicting load weight in lifting tasks.
• The findings provide valuable insights into how hand force is exerted during lifting.
• This technology has the potential to enhance occupational safety by improving risk assessment for manual handling.