Innovating construction through manufacturing: a reinforcement learning framework for robotically-supported modular factories

Principal investigator: Daniel Cardoso Llach and Jean Oh

University: Carnegie Mellon University

Industry partner: iBUILT Group

Representing 13% of the global GDP, construction is the world’s largest industry. However, it is also one of the least automated and most pollutant, with construction sites generating enormous quantities of material waste, dust, noise, water consumption, and carbon emissions. By relocating many tasks from construction sites into factories, where complex components can be manufactured under relatively controlled conditions, modular construction methods have shown great potential to improve efficiency and quality in the building industry while helping reduce its environmental impacts. However, modular factories for building construction rely on a highly diverse and specialized technical workforce whose roles are not all amenable to technical formalization or automation. In collaboration with iBUILT Group, an industry leader in modular construction, our team of computational design, robotics, and architecture specialists at Carnegie Mellon University will develop and test a computational framework for networks of smaller, more adaptive robots to flexibly assist manufacturing workflows within modular factories. Building on our previous research applying state-of-the-art reinforcement learning methods to controlling multiple robots in construction tasks, we will develop a computational framework for logistics and transportation that facilitates the delivery of materials and resources to running manufacturing pipelines at modular construction factories by a multi-robot carrier fleet. iBUILT's existing manufacturing workflows at the company’s factory in Berwick, PA, will be studied to adapt the framework to industry’s constraints. Through simulations and a proof of concept prototype comprising two rovers deployed in the factory we will assess the potential of reinforcement learning methods for robotically-assisted manufacturing to increase both efficiency and flexibility in the manufacturing of volumetric building components, helping reduce barriers to the adoption of modular methods by the architecture, engineering, and construction (AEC) industry, and fostering the emergence of a manufacturing workforce for building construction at the intersection of artificial intelligence, robotics, and architecture.