Predicting Collaboration Requests in Dynamic Construction Environments for Smooth Worker-Robot Collaboration

Worker-Robot Collaboration (WRC) holds great potential in boosting productivity and safety in construction by leveraging the strengths of humans and robots. One crucial aspect of WRC is the ability of assistive robots to anticipate collaboration requests, enabling them to proactively support their human counterparts. However, dynamic construction environments present unique challenges due to constantly changing workplaces, worker interactions, and intentions. To address this challenge, our study focuses on determining when workers require assistance in such dynamic environments. We propose a vision-based approach, employing YOWOv2-based action detection for real-time prediction of workers' locations and their assistance intentions. We evaluate the effectiveness of our method through experiments conducted on a scaffold assembly task, showcasing its efficiency (52 FPS) and accuracy (0.92 frame mAP). These findings contribute to the development of robot assistive policies tailored to dynamic workplaces, paving the way for smooth WRC in construction.