A data-driven approach to detecting human fatigue for adaptation in human-robot collaboration

Lambay, Arsalan Jaweed 2025. A data-driven approach to detecting human fatigue for adaptation in human-robot collaboration. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (6MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (145kB)

Abstract

This research presents a novel data-driven methodology for detecting and monitoring human physical fatigue to support robots in adapting their strategies based on the human physical fatigue state. The detrimental effects of ignoring fatigue in human factors can lead to severe health, socioeconomic issues, and low morale, resulting in institutional and personal losses. Given the subjective nature of fatigue, its quantification is challenging but essential for mitigating its negative impacts. This thesis addresses three primary research questions to enhance fatigue detection and monitoring in human-robot collaboration (HRC) through advanced data analytics and machine learning models. The initial phase of the research involved a systematic literature review of existing methodologies, techniques, and technologies for human fatigue detection and monitoring. It also included an exploration of human-robot collaboration adaptation strategies. The literature review was conducted with three main objectives: understanding the mechanism of fatigue development during physical activity, reviewing state-of-the-art analysis techniques, and categorising these techniques into qualitative and quantitative fatigue analyses with their industrial applications. The integration of machine learning with these analyses has shown promise in predicting fatigue accurately and also highlighting the research gap. Additionally, advancements in sensing technology have enabled the collection of crucial individual information and characteristics through multidimensional and multimodal data. The literature review also highlighted challenges in HRC within the context of the Industrial Revolution 5.0, where increasing human-robot collaboration in strenuous tasks requires innovative fatigue detection methods. Following the literature review, a general framework for the research was introduced, forming the basis for addressing the research questions. The first research question explored how to collaboratively expand the variety of collected samples to improve the detection of human fatigue levels. This involved investigating various approaches such as multisensory, modelling, and synthetic data approaches. Preliminary experimental analysis revealed limitations, leading to the development of a methodology for generating artificial data to capture intricate fatigue details. This helps in answering the first research question of how the data can be expanded to detect fatigue accurately in Chapter 4. This methodology included three steps: identifying input conditions from real data parameters, training a generator with these conditions and controlled noise vectors, and training a discriminator to evaluate the generated samples. Training machine learning algorithms on this synthetic data further enhanced the fatigue detection. This approach allows researchers to systematically generate synthetic data, addressing data limitations and expanding the variety of samples for fatigue detection. The second research question examined the employment of a data-driven machine-learning model to predict human fatigue in real time. Chapter 5, An RNN-LSTM model was proposed to address this, beginning with standard data processing methods and feature selection approaches using LASSO and best-subset methods. These selected features facilitated faster processing time for real-time predictions. Rigorous testing of the proposed model demonstrated its effectiveness in predicting fatigue in real-time, with a balance of features to understand the root causes of fatigue. The second and third research questions are interlinked, with insights from the second question informing the third. The third research question investigated how a robot could propose and implement a dynamically adaptive mechanism tailored to human-robot collaboration when detecting human fatigue. The proposed methodology in Chapter 6 involved using online incremental learning in a Markov decision process for reinforcement learning applied to a use case of surface grinding in an HRC task. This method defines the state, action space, reward, learning policy, and adaptation mechanisms. The features generated from the second research question were utilised to enable the robot to learn and detect human physical fatigue, allowing it to adapt to human fatigue levels in an industrial setting dynamically. The proposed methodology was simulated using ROS2 and Gazebo. Finally, this research contributes a comprehensive approach to detecting and monitoring human physical fatigue using data-driven methodologies and machine learning. The findings support the development of adaptive human-robot collaboration strategies, addressing fatigue detection challenges and enhancing industrial operations' efficiency and safety.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Engineering
Uncontrolled Keywords:	1. Human Physical Fatigue 2. Human-Robot Collaboration (HRC) 3. Machine Learning 4. RNN-LSTM 5. Synthetic Data Generation 6. Reinforcement Learning
Date of First Compliant Deposit:	19 May 2026
Last Modified:	19 May 2026 14:00
URI:	https://orca.cardiff.ac.uk/id/eprint/186759

Actions (repository staff only)

Edit Item