Reinforcement Learning for Design

Over the past few years, the information revolution has changed the face of every industry thanks to an increasing number of sophisticated technological tools that are used for planning and managing projects. In construction, these tools have crystalized as a paradigm that came to be known as building information modeling (BIM), which from a practical perspective can be viewed as a framework for implementing a systems approach to construction projects. In this framework, it becomes possible to capture the global effect of what may seem as the most insignificant local change that is introduced by any given stakeholder, e.g., the owner, the designer, the contractor, etc. However, although BIM allows a variety of models to interact in order for practitioners to gain a global insight into the project dynamics, the design of the first layer of information, i.e., the 3D geometric layout, is still fully dependent on the level of creativity and expertise of the designer. In this paper, we propose to explore the challenges and benefits of using reinforcement learning for designing simple structures. Given that, design is a complex activity that needs to adhere to strict rules of the building code and municipal bylaws, the systems chosen as case studies in this work will focus more on developing simple penalty and reward rules rather than realism.