In an increasingly populated world, civil engineers, architects, and urban planners must intrinsically account for the ergonomics of an environment with respect to the presence, occupancy, and movement of the inhabitants who utilize these spaces. The space of permissible building designs is extremely high dimensional and continuous, and it is infeasible for teams of human experts to exhaustively explore and evaluate possible designs manually, while accounting for non-intuitive and difficult-to interpret factors related to crowd dynamics. Conducting experiments with physical prototypes of environments and real human subjects is simply impossible. Architects, therefore currently rely on their intuition and expertise to make design decisions that are guided by years of experience and practice, but limited empirical evidence. Computer-aided design (CAD) tools offer the promise of automation to predictively analyze, evaluate, and even optimize environment designs. Before we can leverage computation to account for human crowd movement in the building design process, we need accurate models of human crowd movement, and meaningful measures that quantify the relationship between an environment layout and the behavior of its occupants. The far-reaching goal is to harness computational resources to empower architects by efficiently exploring, analyzing, and filtering the design space of possible environment configurations to better inform a designer’s decision-making. Our research focuses on the  algorithms and technologies needed for the computer-assisted design of environments that intrinsically account for the movement and occupancy of its inhabitants.  There are several key research challenges towards meeting this goal, and addressing these challenges is the focus of this research.

User-in-the-loop Crowd-Aware Diversity Optimization of Environment Designs

It is prohibitive to exhaustively explore and evaluate the design space of environment layouts with respect to the dynamic characteristics of crowd movement. It is critically important for automation tools to empower designers (not replace them), by providing multiple, diverse suggestions that meet desired characteristics (e.g., maximizing the accessiblity of important areas such as exit passages) while preserving the original design intent.  The synthesis of multiple, diverse, yet optimal layouts, translates to an extremely high-dimensional, non-convex optimization problem, with competing objective terms, where computationally expensive operations need to be performed within the optimization loop.  

To address these challenges, we have been systematically investigating the relationship between an environment and crowd behavior. Our research deliverables converge towards a computer-assisted solution,  “Crowd-Optimized Design of Environments” (CODE), that optimizes environment layouts with respect to the presence, occupancy, and behavior of human inhabitants. The aforementioned metrics (encoded as deep networks) serve as objectives to formulate and efficiently solve a high-dimensional, continuous, multi-objective optimization problem, to generate a set of diverse, optimal reconfigurations of the  designer’s layout that maximize crowd properties, while preserving the original intent of the designer. Instead of providing a single optimal reconfiguration of the environment layout, CODE returns a set of diverse, yet optimal design alterations, as suggestions to a designer for selection, and iterative refinement. This keeps the user central to the design loop, while leveraging computational intelligence for informed decision-making.   A key novelty of our approach is that instead of simply solving for a single optimal configuration, we solve for a set of diverse candidate solutions.  Our formulation introduces a diversity term in the objective formulation. This requires the solver to focus the search to meet optimality criteria, while simultaneously broadening its exploration to maximize diversity of its candidate solutions. The process of balancing multiple objectives during optimization is a well known challenge, which is rendered even more difficult by the presence of a diversity term.  To address this issue, we have developed a novel hierarchical multi-objective optimization algorithm which balances optimality and diversity while remaining efficient for interactive use.

Coupling crowd modeling within computer-aided environment design is potentially transformative and can enhance architectural and urban design practices, enable real-time crowd management, disaster and security applications, as well as the design of smart cities.