Collaborating with Agents in Architectural Design and Decision-Making Process: Top-Down and Bottom-Up Case Studies Using Reinforcement Learning

Abstract

This study focuses on how architectural designers, engineers, and academics can collaborate with computational intelligent agents in a design and decision-making process, which is a great challenge. Focusing on this idea, a novel approach is presented where designers can use intelligent agents to their advantage for exploring possibilities via data generation and data processing. The problem of collaboration is presented in two distinct approaches—top-down and bottom-up. In the top-down approach, a case is selected where the designer intends to solve a housing design problem starting from meeting the general requirements of total area and distribution of housing unit types. In the bottom-up approach, a case is selected where the designer plans for the very same problem by meeting the specific requirements of room area and relations. Both cases are based on a reinforcement learning (RL) approach in which the user is allowed to collaborate with the RL algorithm, and results are compared both with widely used algorithms for similar problems (genetic algorithms) and ground-truth (deterministic solutions by designer). Compared results of top-down and bottom-up approaches have shown that the reinforcement learning approach can be used as an intelligent data system to explore design space to find an optimal set of solutions within the objective space. Finally, both approaches are discussed from a broader perspective of how designers, engineers, and academics can collaborate with agents throughout the design and decision-making processes.