Worker-centric human-robot co-adaptation in construction

To tackle the productivity, safety, and labor shortage issues, the construction industry has started robotization at various scales in various construction operations. While the deployed robotic solutions have shown the potential to evolve the construction industries into more advanced and productive sectors, most of these efforts cannot advance from the prototypic stage to the practical application in construction. Without workers' supervision and collaboration, the robots alone cannot efficiently execute construction projects in dynamic and cluttered construction environments. To deal with this challenge, current investigations have endeavored to introduce human-robot collaboration (HRC) solutions to help robots perform construction tasks more efficiently and dexterously by taking advantage of human flexibility and versatility. However, bringing collaborative robots to construction sites can raise new concerns regarding workers' safety. These potential safety concerns can be categorized into two groups: (1) physical safety issues – the lack of mutual understanding between robots and workers in terms of locations and movement intentions can lead to a physical collision between workers and robots; and (2) psychological and physiological safety issues – the unparalleled working performance of robots may bring high cognitive load and physical fatigue to workers. To address these safety concerns and thus enhance the reliability of the HRC in construction, this research aims to establish a co-adaptation mechanism that allows robots to perceive their human partners' physical, physiological, and psychological information and co-adapt to their behaviors accordingly. The research carried out three main phases: First, the research developed a vision-based safety monitoring system to enable robots to avoid physical collisions with workers during HRC. The system involves two novel deep networks and a probabilistic collision-checking mechanism to allow robots to anticipate the motions of workers and estimate the collision probability. Second, the research designed a physiological computing mechanism to allow the robot to evaluate workers' physiological (physical fatigue) and psychological states (cognitive load). The mechanism leveraged wearable sensors to capture workers' physiological signals. The captured signals were then processed to train ensemble learning classifiers to distinguish workers' states when collaborating with robots. Third, the research designed a robot behavior model to enable the robot to safely adjust its behaviors (collision-free trajectory, suitable working speed) according to the physical (phase 1) and psychological information (phase 2) of workers. The feasibility of the studies proposed in all three phases was tested by the bricklaying HRC experiment. The experimental results showed that the robot behavior model allowed the robot to use the information of the workers’ motion intention, as well as the collision probability with workers, to re-plan and generate the collision-free trajectories during HRC. Additionally, the model adjusted the bricklaying speed of robots as per workers' adverse physiological and psychological states. As such, the research established a reliable worker-aware human-robot coadaptation mechanism that can allow workers and robots to collaborate with a high level of safety and reliability to execute construction tasks. Furthermore, the findings can make an important contribution to facilitating the dependable establishment of HRC solutions and the safe implementation of robots in the construction industry.