Housing Agency: An Interactive Multi-Objective Satisficing System for the Design of Houses

Co-authored with Kyle Steinfeld. Written as the central paper of the Housing Agency thesis, UC Berkeley, Spring 2012.

Abstract#

The majority of houses built in the United States are designed to satisfy a narrow set of production-efficiency and economic constraints, while essentially ignoring those aligned with environmental performance and client satisfaction. This narrow set of constraints, effectively advocated for by production-side interests, favors design drivers that ultimately lead to mass duplication. While minimal deviations from a pre-determined design pattern are permitted, the production model in place for the vast majority of American homes leaves very little room for anything other than superficial design variation. Lost in this system are the voices of important stakeholders in the design of the suburban built environment — a situation that persists despite the more than 50,000 currently unemployed architects¹ willing to provide design services that would more effectively customize these houses to better suit their occupants, and more effectively optimize these designs to meet environmental criteria.

This paper speculates on the value of applying generative design methods to the production of suburban single-family homes. When seen as a multi-objective design satisficing problem,² we assert that the societal and disciplinary crisis of housing production may be addressed with a higher degree of success in the eyes of each of the competing stakeholders. If a taxonomic framework is developed that allows architects, engineers, planners, and clients to express their often-conflicting concerns in a unified format, then a large number of highly customized, high-performing houses that more successfully meet aesthetic and functional guidelines can be produced, while simultaneously bringing into the process the expertise of more independent architects largely denied access to the current system.

Presented here is an outline of such a taxonomic framework, a discussion of the roles that each level plays in the negotiation of stakeholder values, and a proposal for how a generative design system called the Housing Agency System (HAS) may ultimately be developed based upon this framework. While other generative design systems have been developed that successfully address limited subsets of the problem addressed here — such as single-objective optimizations of the optimal placement of glazing, and multi-objective tools that allow multiple parties to guide a parametric model toward optimal solutions³ — an outline for an open framework that incorporates all the stakeholders involved in housing production remains an unmet challenge.

Introduction#

This paper lays out a taxonomic framework for the proposed Housing Agency System (HAS) — an accretive system of parametric models, analytic routines, and evaluative structures, as well as a set of conventions that governs their use and interaction with design stakeholders. The HAS encapsulates modular contributions from architects, engineers, planners, and clients so as to encourage growth and development over time. The system is divided into three main functional units: Search Constructors, which define a parametric model that acts as a search space for the optimization routine; Simulations, which define processes that are performed on the model to produce metrics to be evaluated; and Goal Sets, which evaluate results from simulations and queries to guide a heuristic solver towards a more satisfactory result.

Below, we present a summary of each of the elements that make up the taxonomic framework and the procedures involved in orchestrating these elements in order to generate design options. As an illustrative example, sample stakeholder interactions with the HAS are then presented, using the familiar case of the Villa Savoye by Le Corbusier. The Villa Savoye is then "mutated" by adjusting the search constructors and goal sets based on the needs of the family building the house, the planning commission of the new development, and modern construction methods.

Search Constructors#

In the HAS, Search Constructors collectively play the role of a "design space" that represents the range of possible design solutions at any given time in the design process, at five discrete architectural scales. Search Constructors allow design stakeholders to generate new design iterations. Computationally, each search constructor is described as modular sets of parametric relationships defined collectively by design stakeholders. These interact to form a robust parametric model, or Search Model. When implemented on a design project, each search model requires description at five architectural scales to be complete: Planning Strategies, Formal Strategies, Construction Systems, Surface/Detail Systems, and Building Components. The resulting parametric model may be manipulated or evaluated by other actors in the HAS in order to generate a range of valid design options or to quantify the success of specific configurations.

Planning Strategies include the planning for one or more units simultaneously, and account for interaction between the various units and their impact on the larger community.

Formal Strategies are 3-dimensional massing organizations that include constraints which can be customized based on goal sets to limit the search space. Ranch, Cape Cod, Sea Ranch, and Savoye House are examples of existing formal strategies. Interior Formal Strategies, a subset of formal strategies, respond to typical conditions that are found in formal strategies and allow for a further level of customization to Formal Strategies if the author allows.

Construction Systems rationalize the construction of formal strategies, and allow material scientists, architects, and engineers to go to market with products by advertising them on the marketplace supported by the Housing Agency system. Light Frame, Heavy Timber, Light Gauge Steel, and Rammed Earth are examples of existing construction systems.

Surface / Detail Systems augment construction systems and are adaptable for varying climates, seismic conditions, and assembly requirements. Rain Screen, Hurricane Clips, and Board-and-Batten Siding are examples of existing surface/detail systems.

Building Components are pre-fabricated or manufactured elements that alter the construction system. Windows, Doors, Stairs, and Mechanical Equipment are a few of the many examples of building components. They are built to interact with construction systems, and may be custom fabricated or chosen from a database of commercially available products. The position of building components may be manipulated by the solver — allowing the design assistant to evaluate options for an ideal daylighting scenario for the client's activities.

Simulations#

In the HAS, Simulations play the role of quantitative evaluations that help narrow defined search spaces and inform the definition of goal sets. Simulations allow stakeholders to evaluate existing design iterations and to set performance metrics to which these iterations must comply. Computationally, each simulation is described as a procedural query performed on design iterations that returns evaluative information as a decision assistant, or to guide a solver towards a desired result.

Structural Simulations evaluate a geometrical configuration in combination with a construction system to determine if the configuration meets minimal requirements for gravity, wind, and seismic loading.

Cost and Schedule Simulations evaluate the first and approximate lifecycle costs of a building. The material can be evaluated as well as labor costs associated with a process. Construction systems that specify their processes by contractors with historical data or verified resources may also include a time-linked phasing parameter to generate schedules.

View and Sound Simulations use sensors dispatched to a specific site, as well as feedback from human agents, to determine the STC rating of a wall assembly or the placement of axes or glazing to reinforce or exclude a specific view.

Geometrical Queries implement information such as the volume, area, or dimensions of a space, as well as relational factors such as proximity or overlap, as potential queries.

Simulations play a key role in the generative cycle of the system: simulation results are gathered from the analysis of a defined Search Model and then evaluated by Goal Sets. The HAS then determines the action to be taken in order to manipulate the Search Model to better satisfy the various Goal Sets. The simulations described above represent a comprehensive list of those that have been successfully integrated into the HAS. Building scientists, engineers, contractors, or other qualified stakeholders may contribute additional simulations in the future.

Goal Sets#

In the HAS, Goal Sets quantify stakeholder values. They are described as specific numeric targets, ranges, or benchmarks that determine the degree of success that a particular manipulation of the model has produced from the point of view of a given client, site, or design team. Goal Sets often conflict, as a planner may have a benchmark that conflicts with the ideas of the homeowner and architect. This productive tension creates dynamic environments that avoid the strict homogeneity of traditional zoned developments and more accurately represent the negotiation of values inherent to intervention in the built environment.

Designer Goal Sets: Designers declare the importance of characteristics of projects through their Goal Sets. Existing systems for evaluation such as the International Building Code or the LEED system are considered Goal Sets.

Community Goal Sets: A planning commission can set up a Goal Set which can be opened up to the members of a locality for collaborative design. The production of community goal sets may define a formal order and minimum performance benchmarks that must be adhered to in order to build.

Client Goal Sets: Clients provide goal sets based on surveys or interviews with architects to determine the specific needs of the future homeowners. The number and type of rooms in the home, as well as the lighting, sound, and cost requirements, are examples of parameters that may be modified within the Client Goal Set.

Goal Sets are declared by the various stakeholders to provide metrics for judging the satisfaction of schemes developed from the Search Model. This allows a vast design space to be explored, judged, and guided recursively using a heuristic solver guided by Goal Sets.

Stakeholder Interaction#

This section presents views of the system from the perspective of the various stakeholders. The case study described below uses the Villa Savoye by Le Corbusier as a vehicle for explaining the various methods of interaction, and illustrating the process through which Designer Goal Sets, Planning Strategies, Formal Strategies, and Construction Systems are defined and negotiated.

Designer Profile#

The Designer Profile includes the collection of system elements created by an architectural designer. In the hypothetical scenario described below, a profile representing Le Corbusier — based upon Five Points as well as "The Problem of the House" in his 1923 manifesto Toward an Architecture⁴ — is entered into the system. This profile includes a ranking of required values as well as conditional statements that must be satisfied in order to use the Corbusian goal set.

Planning Strategy Profile#

The Savoye House Planning Strategy is acquired from analysis of the existing building, determining the metrics used in placement and orientation of the house. To demonstrate the Planning Strategy, a community of houses that follow the Savoye House formal strategy and designer goal set are generated using a variety of Client Goal Sets.

Formal Strategy Profile#

In our hypothetical scenario, Le Corbusier adds projects completed in the past to the Housing Agency system. He creates a Formal Strategy model to articulate the form of the Savoye, constrained by some fixed elements that he wishes to keep in all iterations of the Formal Strategy — such as the ramping circulation, raised second floor, and proportion of the roof terrace. He also creates Building Components including the ramp, ribbon windows, and a spiraling staircase that are flexible to respond to manipulation of the model.

Construction System Profile#

Le Corbusier also creates a construction system for the Villa Savoye called the Pilotis. The construction system takes advantage of hollow steel-section columns above pier foundations, plaster-coated masonry walls, and reinforced concrete floor slabs with member sizing, rebar spacing, and configuration determined by structural simulation output. This system uses a rigid core that takes lateral loads, allowing for thin, unbraced gravity-supporting columns at the periphery.

Client Profiles#

Client Profiles are generated by clients in order to constrain the set of candidate solutions to those that are acceptable to the client. The generation of a client goal set is assisted by a survey. In our scenario, the Persimmons family want to build a home that has to support the needs of their family as well as John and Tina's aging parents, who require on-site care. They require a fully accessible home that has an open floor plan to support frequent gatherings. John appreciates the clean lines and proportions of modern architecture, but he is a green-technology advocate and wants to keep his house as compact and energy efficient as possible.

Conclusions and Future Work#

The taxonomic framework described in this paper allows for the recording of processes typical in the production of houses. The modular and flexible nature of the generative design system allows for a vast database of responses to accumulate, and the defined goal sets allow those responses to be customized to the environment, community, and client. Compared to the existing method of housing production, where mass duplication is the norm, the HAS allows for the design community and those purchasing homes to play an active role in the production of houses.

Future work on the HAS includes creating modeling and naming conventions that allow the system to become modular. A series of components for Grasshopper and Rhinoceros are in development which allow for assistance in the creation of Formal Strategies, Building Components, and Planning Strategies. Implementation in a Building Information Modeling platform is planned as well, as the "Family" model of parametric components and a common object-oriented programming environment presents an ideal implementation environment for the HAS.

References#

Clark, C. J. "Exactly How Many Architects in the U.S. Are Unemployed?" Architectural Record, 10 Oct. 2010. Web. 21 Apr. 2012.
Holzer, D., and Downing, S. (2010). "Optioneering: A New Basis for Engagement Between Architects and Their Collaborators." Architectural Design, 80: 60–63. doi:10.1002/ad.1107
Negroponte, Nicholas. Introduction. Soft Architecture Machines. Cambridge, MA: MIT Press, 1975.
Le Corbusier, and Frederick Etchells. Towards a New Architecture. London: Architectural Press, 1946.
Morel-Journel, Guillemette. Villa Savoye. Paris: Éditions du patrimoine, 1998. ISBN 2858221952.

Clark, C. J. "Exactly How Many Architects in the U.S. Are Unemployed?" Architectural Record, 10 Oct. 2010. ↩
On satisficing as distinct from optimizing, see Herbert Simon's work on bounded rationality. ↩
See Holzer, D., and Downing, S. (2010), "Optioneering: A New Basis for Engagement Between Architects and Their Collaborators," Architectural Design, 80: 60–63. ↩
Le Corbusier, Toward an Architecture (1923); quoted here in the Frederick Etchells translation, Towards a New Architecture (1946). ↩