Worker Fatigue Detection for Interoperable Co-Robot Safety in Construction

This project develops a methodology for detecting fatigue in workers that is robust to worker identity and task variability in construction sites. This information allows us to assess the efficacy of co-robots to mitigate fatigue-related injuries. This project integrates wearable sensing and machine learning methodologies with current ergonomics and safety models. The long-term goal of the research is a data-driven safety model that provides real-time assessment of a worker’s condition, links safety to performance, and optimizes co-robotic interventions to worker fatigue. The short-term objectives are developing and validating a change-point based detection methodology for use in worker evaluation in unstructured construction environments. 

We are exploiting the repetitive nature of typical construction tasks and the prior knowledge that certain tasks will result in fatigue. The problem then turns into estimating the time of onset and potentially the severity in a way that does not require re-calibration between users and general task classes. We are using online change-point detection on weakly informative signals, i.e. where changes are both gradual and subtle. The method can be used online for a variety of tasks and sensors. We have formed an interdisciplinary team of experts from ergonomics/safety, robotics/machine learning, and the construction industry to implement and validate such a system. The activities advance both our knowledge of how real-world co-robotic systems affect fatigue and add a new tool for designing co-robotic systems to effectively mitigate fatigue and making workplaces safer.

The outcomes from this research include an interoperable, data-driven, system to provide real-time assessment of likely fatigue state. This includes both the method of collecting data (PI Cavuoto), and an algorithm to process the data in real-time (PI Napp). This method will be validated with one of the SMART’s industrial partners (Construction Robotics) and provide valuable real-world data about how their co-robotic products affect worker fatigue. The long-term impact will be establishing SMART as a center for co-robotic safety decision making and fatigue detection expertise in the greater region. This project falls within the SMART themes of humans-in-the-loop and information processing, while building upon an industry partnership.

For additional information on this project, please contact either Dr. Lora Cavuoto (loracavu@buffalo.edu) or Dr. Nils Napp (nnapp@buffalo.edu).