HappyValley Industry recently reported on the groundbreaking research being led by a team of engineers at Penn State University on human-robot collaboration on construction sites. Their research has shown that the development of a collaborative robotic system that understands the human partner’s mental state could improve workers’ safety and productivity in the long term.
HappyValley Industry had the opportunity to connect with Penn State University Assistant Professor of Architectural Engineering, Dr. Houtan Jebelli. Dr. Jebelli also serves as director of the Construction Robotics, Automation, and Intelligent Sensing Lab (RAISE Lab). In our Q&A, Dr. Jebelli further discusses his team’s work in the construction robotics field and covers in-depth the concept, development, effectiveness and future of construction automation and robotics.
HappyValley Industry: How did the concept of using robots on construction sites come about?
Dr. Jebelli: The construction industry suffers from stagnant productivity, labor shortages and unsafe working conditions. Such impediments severely affect the well-being of the construction workforce and productivity growth in this $10 trillion-dollar global market. The construction productivity has trailed that of other industrial sectors for decades, with 1% labor-productivity growth compared with 2.8% for the total world economy and 3.6% for manufacturing. Notably, construction-related spending accounts for 13% of the world's GDP while the industry's annual productivity remains relatively constant – a 1% increase – over the past 20 years in the last 20 years. Unfortunately, the construction industry falls behind many other sectors with respect to occupational safety and health. Close to 20% of annual fatal injuries are in construction, the second largest number of deaths after transportation and material moving occupations. In 2019, 1,066 workers died at construction sites. The aging workforce is also a big concern. The number of American workers age 55 and over has more than doubled from 1992 to 2017
"The construction productivity has trailed that of other industrial sectors for decades, with 1% labor-productivity growth compared with 2.8% for the total world economy and 3.6% for manufacturing."
Construction automation and robotics is a response to these detrimental variables, a major thrust towards industrialization. This is a paradigmatic shift that attempts to transform the industry from a stagnant, conservative sector into an automized one. This transformation has become possible due to the advent of sensing technologies, computer processing powers and artificial intelligence. These advances will pave the way for adopting intelligent machines, autonomous vehicles and perceptive cyber-physical systems at construction sites.
HappyValley Industry: Can you explain a little bit about how the robots can be used effectively on the construction site?
Dr. Jebelli: Most of the manual and strenuous activities in construction are repetitive tasks that can be performed, at least in part, using construction robots, such as brick-laying tasks, tying rebars, painting walls and even welding structural elements. There are also tasks wherein the field manager needs to ambulatory inspect the site for monitoring construction progress or inspecting workers' health and safety. With the advent of unmanned aerial and ground vehicles, these tasks can be transformed into an automated process with more systematical evaluations. Several other examples can show the potentials of robotic systems to deliver activities at the construction job sites, such as constructing concrete walls using 3D-printed robots or reconnaissance operations using unmanned terrestrial vehicles.
"This is a paradigmatic shift that attempts to transform the industry from a stagnant, conservative sector into an automized one."
The essential point to contemplate is that many of these activities require human dexterity or "a finishing touch" for appropriate completion. While these robots can provide an excellent opportunity for construction firms, without workers' management and help, these robots cannot deliver the job as expected, specifically in the dynamic environment of construction job sites.
HappyValley Industry: Monitoring the brainwaves of the construction workers is very interesting; can you talk about how that data will be incorporated into the activity of the robot?
Dr. Jebelli: One of the key points in worker-robot interaction is the ability of the robot to recognize its human partner, not as another generic element in the scene but a biological partner that can be fallible, fatigable and irrational. Without a communicative channel between the two peers, robots may impose substantial cognitive and physical stress on the worker. To alleviate this situation, we propose a brainwave-driven framework that extracts the data from the worker's non-invasive wearable sensors, applies advanced machine learning techniques to translate these signals into meaningful mental states and leverages a robotic operating system to transform the interpreted states into robotic commands. This will allow the robot to adjust its performance based on the mental workload of the workers.
HappyValley Industry: To that point, what have been the biggest successes of the study and what have been the biggest challenges?
Dr. Jebelli: The biggest successes of this study are twofold: 1) the successful interpretation of workers' mental workload from the EEG signals of the brain; and 2) enabling the robot to successfully adjust its performance and action based on the interpreted, discrete mental states.
"Our proposed work on human-centered worker-robot collaboration has demonstrated the feasibility of worker-robot adaptation based on the real-time stream of physiological signals captured from workers' wearable biosensors. Yet, this is just a start."
The biggest challenges of this research were 1) the artifact removal and 2) interpretation of the brainwaves. The EEG signals contain intrinsic and extrinsic artifacts (environmental noise, body movement and ocular artifacts) that can substantially reduce the quality of the signal. We were able to overcome this challenge by de-noising the signals using several effective filtering techniques such as the fixed-gain filtering method, blink-source separation technique and dependent component analysis (DCA). The interpretation of the signal was challenging because of the inter-subject variability, an important phenomenon that explains why different workers react differently even under the same working condition. To alleviate this problem, we leveraged ensemble learning techniques with the ability to reduce generalization errors in our data-driven models and improve the robustness of the model in predicting various mental states of the workers.
HappyValley Industry: What is the next step?
Dr. Jebelli: Our proposed work on human-centered worker-robot collaboration has demonstrated the feasibility of worker-robot adaptation based on the real-time stream of physiological signals captured from workers' wearable biosensors. Yet, this is just a start. We need further research to enhance the reliability of the framework in other field-oriented worker-robot collaborations by performing diversity sampling, increasing the number of subjects and improving the prediction accuracy of the data-driven models. The incorporation of other contributory variables into data analysis and interpretation can profoundly improve the accuracy and robustness of the embedded translative pipelines, such as including environmental and contextual variables. Another critical area that needs a great deal of attention is the sensitivity analysis of the framework to realize how other physical and physiological conditions will affect the mental states of the workers and how the interplay between these mental and physical states can be realized to further the precision accuracy of the framework.