A Human-Centered, Reinforcement Learning Driven Robotic Framework For Assisting Construction Workers

This research explores how robots can be designed to support, rather than replace, labor-intensive workers in construction. It develops a robotic framework leveraging Reinforcement Learning (RL) algorithms, encompassing robot hardware prototyping, unstructured site perception, worker detection and tracking, hierarchical motion planning, and contextualized RL training. Realized as a “work companion robot” for carpentry workers, the framework is tested in both lab and real construction sites. Key contributions include a practical RL-driven robot for tool/material delivery and worker-robot interaction, a modular and lightweight system with tailored control and sensory packages, an comprehensive perception architecture for navigating unstructured environments, a hierarchical motion planning stack that integrates RL-based and search-based path planning, and a contextual RL finetuning pipeline for social navigation around workers. This robotic framework lays a technical foundation for future research into robotically-supported collaborative work, advancing the field of computational design, robotics, and AI/ML by integrating advanced robotics into real-world construction contexts with a focus on human-centered support.