top of page
Unleashing the Power of AI to Transform
Ergonomics and Improve Workplace Safety

Revolutionizing
Ergonomics

 

'AI-driven Ergonomics Solutions for Safer Work Environments'

Welcome to the future of ergonomics! At PIKUP, we blend cutting-edge AI technology with ergonomic expertise to revolutionize workplace safety and efficiency.

Based in Arizona State University, our dynamic team of AI and ergonomics experts is dedicated to developing innovative computer vision software that detects and corrects bad postures in lifting tasks.

Join us in reshaping the way we work and ensuring a healthier, more productive future for all.

Research Question

How does the implementation of an AI-driven alert system affect workers’ compliance with safe lifting standards and reduce lifting-related Stress?

Project Scope

goal.png

Goal

Increase safety among warehouse workers (avoiding repetitive and potentially harmful actions) 

setting.png

Environment

The Environment setting takes place at a Warehouse/ Factory

clipboard.png

Task/Tool

An AI/Vision system that monitors worker actions and issues an alert when it detects such dangerous actions (Bending over to lift boxes instead of squatting). The tool can send recommendations for retraining when the activity is done repeatedly.

Problem Statement

Screenshot 2024-04-14 at 9_edited.png

Our proposed solution is to bring in a camera-based real-time detection and alerting of dangerous postures.

Hypothesis

hike.png
Increase Compliance
109613.png
First
Response
solution.png
Reduced
Stress

Study Design

Design: 

 Between-subjects experiment.

Two different groups:

Group A started with software and had it removed

Group B lifted first before software was introduced

Variables: 

Lifting with Software / Without software

Measurement: 

Compliance with safe Lifting, Self-Reported Stress Level

Participant's Information

● 10  participants

● Age Range 18-34

● All Male (unfortunately)

● 70% performed lifting tasks frequently

● Most lifting tasks performed at home

● All familiar with the use of AI

 

Design Experiment

• Participants provided informed consent before the start of the experiment (Information sheets were given to each participant)​

• Participants completed lifting tasks with or without software depending on their group.​

• Participants filled questionnaire individually after each task.

Prototype

• Pose model trained on real human images. 

 

• Uses pose information to detect bad posture

• Triggers an alarm when it detects a bad posture with 80 % confidence (Chosen to avoid False Positives)

Experiment Report

83.3%

of participants believe the AI Alert system has positively impacted the work routine

83.4%

of participants found the AI system effective at reminding them to lift 

83.3%

of participants experienced reduced levels of stress and discomfort after using the AI system.

Discussion

Compliance to Safe Lifting
Significant!!
The use of software had an effect on the lifting techniques of participants during and after the use of the software.
Physical Discomfort and Stress
Significant!!
Using the Software reduced stress levels as reported by participants

Limitations

Inability to test on the target group (Factory/ Warehouse workers)

Lack of standard and variable dataset (Our dataset had less variability, which affected the accuracy of our model)

Future Scope

●LONGITUDINAL Study, the effect of using the system over time.

 

●Building more familiarity with the system

●Indicating the exact wrong posture

●Color coding of poses

●Apply the lifting equation directly from pose Landmarks or Use a combination of both (PIML)

●More detailed statistical analysis

Conclusion

  • The AI Alert System can help compliance with safe lifting and reduce physical discomfort and stress levels.

  • A more refined version of our software has the potential to improve safety among people performing different lifting tasks.

 

  • We believe repetitive use of the software will lead to a learning effect, resulting in safe lifting without the software (Longitudinal study can test this)

bottom of page