A Reinforcement Learning Based Approach for Conducting Multiple Tasks using Robots in Virtual Construction Environments

Construction robots are considered a promisingsolution for reducing onsite injuries and increasing productivity.One of the bottlenecks in deploying construction robots issolving the problem of robotic motion planning, consideringthe dynamic nature of construction sites. Specifically, currentworks in robotic motion planning for construction lack thegeneralization capacity for different tasks (i.e., a robot isgenerally optimized for a highly specialized task and fails togeneralize when the task deviates slightly from its originalform). In this paper, we proposed a reinforcement learningbased approach for robotic motion planning using curriculumlearning, which enables robots to conduct multiple constructiontasks using a single trained agent. We tested our approach onthree common construction tasks (ceiling installation, windowinstallation, and flooring), resulting in an average success rateof around 80%.