How to Cite: Fraile-Narvaez, Marcelo. "Using Swarm Intelligence in Urban Design". Dearq no. 37 (2023): 43-53. DOI: https://doi.org/10.18389/dearq37.2023.05
How to Cite: Fraile-Narvaez, Marcelo. "Using Swarm Intelligence in Urban Design". Dearq no. 37 (2023): 43-53. DOI: https://doi.org/10.18389/dearq37.2023.05
Marcelo Fraile-Narvaez
Escuela Superior de Ingeniería y Tecnología.
Universidad Internacional de La Rioja
Received: August 23, 2022 | Accepted: April 10, 2023
As a result of the evolution of digital technologies and information networks, swarm intelligence has come to be implemented in a range of research perspectives. Building on the assumptions that characterize the approach, this paper aims to unpack the importance of swarm-based urban design, which is opening up a new frontier as part of an emerging approach and which lends itself to deployment in myriad situations. This article focuses on a case study of the Kokkugia studio's Melbourne Docklands project. Although this scheme has not been built, it has paved the way for new explorations and ways of understanding urban design.
Keywords: emerging urbanism, swarm urbanism, swarm intelligence, multi-agent systems, self-organizing systems, particle swarm optimization, swarm optimization.
Conceptually, swarm intelligence may be considered a discipline that focuses on the ways a group of autonomous agents are capable of working together without centralized supervision. Essentially, it is based on the idea that the division of labor, self-organization, and adaptation of individual agents can result in more efficient and effective solutions than those provided by a single complex entity.
Swarm intelligence can be applied across a wide variety of fields, including robotics, urban planning and the search for optimal solutions within complex systems. Despite being a relatively new discipline it has made significant progress, driven by developments in digital technology and computer science. With the ever-increasing ability to process and store data, computational algorithms and models have been developed which are capable of simulating and optimizing the behavior of individual agents, thus leading to real-time, complex problem solving and system improvement across a wide range of applications. One of the best-known approaches in this regard is particle swarm optimization, which demonstrates how a set of simple agents can cooperate to find optimal answers to challenges that would be difficult to solve individually. This approach has been a cornerstone of the evolution and understanding of swarm intelligence and has provided a solid foundation for its application in a vast range of fields.
In urban environments, swarm intelligence has been applied to the planning and design of smart and sustainable cities. An interesting example is the planned redesign of the Melbourne Docklands in Australia, by the Kokkugia studio. The firm used swarm intelligence to simulate and optimize the shape, structure and function of buildings, resulting in greater efficiency of spaces and improved sustainability in urban planning processes, which in turn have the potential exert a positive impact on the quality of life of the area's inhabitants.
There is a strict division of labor between the bees that live in a hive: from the workers to the drones to the queen bee, each has a precise role to play, within a perfect natural mechanism. This allows them, in particular, to optimize the search for the flowers they need to produce food, since every colony extends its domain over a vast topographical territory. The first to take part in this process are the scout bees, who are responsible for finding food by searching for the highest concentrations of flowers in an area, a task they perform instinctively, in a refined, random, behavior developed over millions of generations. Once they have found a site, they return to the hive to report their discovery. For this purpose they perform a "dance" at one end of the hive. The workers are attracted to the dance by a sound emitted by the scouts as they perform.
Conceptually, this dance configures a body language (known as bee language), which was discovered in 1919 by the Viennese professor Karl von Frisch. By vibrating their abdomens, scout bees inform the rest of the hive where their source of sustenance is located. Once the show is over, the workers respond to the choreography of the scout that impressed them most, and who indicated where the best food was found.
However, the ingeniousness of the process lies in the complexity of the system, despite the limited intellect of each individual agent. The procedure involves a large number of individuals with minimal cognitive abilities who are capable of solving difficult problems by following a limited number of simples rules. In addition, each bee has a notion of its own position and that of its immediate companions, but -as a part of a swarm made up of hundreds of individuals- interact with the whole and are able to change position in masse when the situation warrants it, in a matter of seconds.
This behavior lies behind the concept of swarm intelligence, a term originally borrowed from the field of biology and which has been adopted by different areas of research to explain the collective behavior of decentralized systems exhibited by animals of similar size. Self-organized structures inspired by "the collective way of acting of very uncomplex societies, composed of very unsophisticated individuals" (Duarte Muñoz, Pantrigo Fernández and Gallego Carrillo 2007, 101).
In nature, there are many other examples of this type of society, including schools of fish, ant colonies and flocks of birds. These are communities that behave as if they were a single individual (Duarte Muñoz, Pantrigo Fernández and Gallego Carrillo 2007). It is an intelligence that emerges from elementary entities with reduced intellect that, as a group, or colony, develop a collective behavior that allows them to solve problems with great creativity, resulting in a superior intelligence which, behaviorally, is greater than the sum of its parts. Unlike human hierarchical structures, which have directors or supervisors, in a swarm there is no individual who governs the system. Its products are a consequence of an emergent property -the chaos concept of scalability- an unplanned operation without a centralized control organization directing its proceedings, involving local interactions between elements that lead to coherent integral behaviors (Duarte Muñoz, Pantrigo Fernández and Gallego Carrillo 2007).
There is a correlation between the individual and independent operation of each part, without there being any central monitoring structure that determines the strategy to be followed. Each component maintains a certain independence, but acts in cooperation with nearby components. As Duarte Muñoz, Pantrigo Fernández and Gallego Carrillo say (2007, 101), "although the agents are simple, the result of their global interaction can be very complex (such as carrying out movements practically in unison, or [developing] protective strategies against predators)".
Currently, as a result of the evolution of digital technology and information networks, swarm intelligence is being implemented in different fields, including biology, social sciences, engineering, computer science and, of course, architecture and urbanism. Building on these assumptions, the main objective of this research is to unravel the importance of urban design based on swarm intelligence. This is a new frontier for research, with a myriad of potential applications, including data mapping, the modelling of collective or individual behavior in an evacuation, or the assembling of elements using nanorobots. The analysis presented in the article is based on a case study focused on the Kokkugia studio's Melbourne Docklands project, a speculative exercise that opened the door to new ways of understanding and approaching urban design in the 21st century.
Finally, the article is intended to create a theoretical framework that will serve as a starting point that will encourage and promote future work and discussion, thereby advancing the topic as a whole.
The literature on swarm intelligence encompasses an enormous variety of themes, somewhat disconnected at times and which covers such dissimilar fields as physics, biology and robotics. In this synthetic chronological survey, we consider the British philosopher and writer William Olaf Stapledon to have been the first to speak of swarm intelligence. In his book Last and First Men: A Story of the Near and Far Future, published in 1930, he described the history of humanity from the present to two billion years in the future. Using a Hegelian dialectic, Stapledon developed the idea of a series of individual cells that communicated with each other through radio waves.
Years later, in 1986, with the popularization of computers, the expert in computer graphics and artificial life, Craig W. Reynolds, developed Boids, a digital system that made it possible to visualize the emergent behavior of a flock of digital birds (boids1). In this model, each agent chose its own course and navigated according to its perception of a dynamic environment that governed its movement, a perception that responded to the three basic rules of swarm behavior. That is: each agent maintained a minimum separation from its neighbors, while also maintaining a directionality that was coherent with that of the flock as a whole, and remaining united with it. Ultimately, the behavior of a swarm of boids was the result of the interaction of individual behaviors: agents tried to fly together, avoid collisions with each other, and moved within the confines of the changing frame that surrounded them.
Later, in 1989, inspired by natural phenomena, the American professors of electrical engineering Gerardo Beni, Susan Hackwood and Jing Wang, introduced the concept of swarm intelligence within the context of mobile robotic systems, to refer to a large (though finite) number of robots with no central control that were in communication with other nearby bots operating in an n-dimensional space. Together, these bots were able to develop local interactions and generate a global procedure that could be used to describe emergent collective behaviors (Beni and Wang 1993). Thus, for Beni and Wang (1993) swarm intelligence may be defined as the property of a system in which the collective behavior of (unsophisticated) agents interacting locally with the environment leads to the emergence of global functional patterns in the system.
Later, in 2013, physicists Maksym Romensky and Vladimir Lobaskin of University College Dublin (Ireland) discovered new collective properties of swarm mechanics while studying dynamic self-organization and order-disorder transition under a two-dimensional system of self-propelled particles. Using digital simulators based on material models, they began to study the sorting parameters of particles relative to their neighbors. They studied the behavior of 10,000 individual digital insects moving at a constant speed on a flat surface. As a result of these studies, Romensky and Lobaskin (2013) were able to assess the behavior of these organisms within a swarm, a behavior that was dependent on the number of individuals, their topographical position within the model and their proximity to other subjects.
The Particle Swarm Optimization process was introduced in 1995 by American scientists Russell Eberhart and James Kennedy, during the course of their exploration of a prototype used to describe the social behavior of animals in group settings. During their research, Eberhart and Kennedy discovered that the model was capable of performing different optimizations. Essentially, it is a numerical variable optimization mechanism, through which cooperation between agents aims to find the global minimum value of a function, i.e. the solution to a problem.
The improvement process begins with a particle population (swarm): a series of uniformly distributed random components whose objective is to obtain an optimal value (response) to a given problem. For this purpose, each particle occupies a position in the search space, which is also the place where possible optimization solutions are found. Each particle is fully aware of everything happening around it (the spatial sphere), and can remember its most favorable spatial position and optimal experience. However, it is unable to observe what is happening outside the spatial sphere.
Each particle moves in space according to its own alignment, to a length vector that travels in an n-dimensional space (the search universe) and displays evolution over time, producing changes in its speed that are required to maintain its distance from other particles found close to it, within its sphere. In addition, the direction in which the flock moves is maintained2.
Likewise, as a swarm, each particle has knowledge of the best historical position of the particles in its vicinity, subsequently providing an additional factor that helps it move towards that location, contributing to the overall experience of each particle. A cohesion principle also exists, whereby the entire swarm moves together as a group. Thus, these three parameters —particle direction, maximum individual experience and superior swarm dexterity— describe the mathematical model that defines the system.
Once the process has been initiated, the system continues to iterate, changing the position of the particles until they locate an area where they converge, randomly, in groups to a true global minimum value. In essence, this is a cooperative process that attempts to find the optimal location within the search universe. That is: to discover the best possible answer to the optimization problem.
In 2001, in his book Emergence: The Connected Lives of Ants, Cities and Software, the American writer Steven Johnson, described cities as a collection of individuals who interact with their neighbors according to the laws of adaptive dynamic systems: a population composed of a large number of small discrete elements, which display sophisticated interactive behaviors in emergent patterns similar to those developed by a colony of ants or a flock of birds. And it is in this sense that, like any emergent structure, a city may be considered to be a pattern in time, displaying an ascending collective intelligence, whose overall sophistication is greater than that of its members. In other words, as a form of "swarm intelligence" (Leach 2009, 56-63). This is a relatively uniform process in which individual components do not stand out but instead conform to the dominant logic of their environment.
And while the complexity of the city far exceeds that of any currently existing digital model, Johnson (2001) continues, the use of computers makes it possible to develop computational methodologies that help us understand the emerging logic behind the ways in which nature self-organizes, and then to use the lessons learnt to simulate collective behaviors within cities and to design urban forms of different size. This is a process that contributes to overcoming approaches inspired by fractals, L-systems, cellular automata and other methods that operate largely according to their own internal discrete logic. It should be borne in mind that both fractals and L-systems are limited in their ability to model growth patterns, as they are programmed to behave in a particular way and, in general, cannot adjust their behavior in response to external stimuli (Leach 2009). Similarly, cellular automata, while capable of responding to their neighbors, are spatially fixed, being tied to certain underlying grids.
This is where swarm intelligence appears to demonstrate its superiority, as it is a multi-agent system composed of different intelligent elements that interact with each other and are capable of moving spatially. It is a dynamic, adaptable, interactive system, able to optimize a shared objective by engaging in a collaborative search, an emerging experimental approach that fuses the biological and the digital and converges in a new scientific and technological field. It is a complex, artificial, method, a generative design process, which uses algorithmic techniques to develop a computational methodology based on swarm intelligence. For Manuel DeLanda, these algorithmic models make it possible to generate virtual multi-agents, which are capable of making their own decisions and influencing the choices of others, a procedure where the emerging autonomy of each individual replaces the notion of centralized control (Leach 2009). Flexible, context-adaptive formats capable of operating at different scales provide the possibility of analyzing, simulating and evaluating multiple options and variations.
This urban analysis employs a large number of virtual entities known as particles. In each system, particles may represent people, buildings, vehicles, roads or public spaces. The particles are distributed across the exploration space, constituting an informal swarm of reference points in the process of building relationships with each other: a colony of elementary nodes, which, like bees, act as an active nucleus that communicates with its peers in real time and follows basic rules: move in the same direction as your neighbors, stay close to them, and avoid collisions. Consequently, each agent follows its immediate neighbors, calculating these rules several times per second, but without any conscious awareness of the overall group.
Thus, the position of each point, determined by coordinates, represents the values taken by the problem's decision variables. Each particle produces a result that is a function of its current position and expected location. In each interaction, the chosen algorithm modifies the situation of the individual, using a velocity vector associated with the particle, which seeks paths to travel along using a set of parameters that represent possible alternative solutions (Duarte Muñoz, Pantrigo Fernández and Gallego Carrillo 2007). These points are transformed into a flexible mesh that supports the inclusion of different types of urban constructions, generating infrastructure and circulation networks: dense, fluid forms that are capable of connecting new structures to existing configurations in a more efficient way.
Within a system that uses swarms for urban planning, the city is no longer made up of static objects: designers consider the city and its buildings as a swarm of interactive installations, involving a process of digital simulation of agents in space, defining and identifying their limitations in order to achieve a balance within the system.
Take the following example: using swarm urban planning models, we might attempt to place 1,000 single-family dwellings in a low-density development area. Initially, the system will be organized in a specific way, certain distances being maintained between individual agents. According to the algorithm, each house will establish its position in relation to its neighbors. This is an open behavior, meaning that if one agent changes its position the system will respond to the new parameter by redefining itself. And given that this type of design permits different swarms to interact simultaneously during the process, it is possible to create a swarm of streets, a swarm of squares and another of public buildings, which will interact with each other until it becomes possible to reproduce the complexity of a contemporary city by digital means.
For Leach (2009), the challenge of this operation is to establish what he calls scenario planning, a procedure that can provide the predictability required to mimic urban processes. That is: the rules that most effectively anticipate procedures capable of generating this level of complexity. Some codes may bring life, others boredom and others still, the death of the hive even. Essentially, this is a trial and error system, operating at high speed, capable of developing millions of possible outcomes and an infinite number of versions or variations (Kievid 2014). And it is precisely here where the designer's task is critical in defining the variables of interaction between agents. The designer must adjust the parameters and look for rules capable of elaborating a balanced structure, making it possible to optimize the swarm and keep the process alive. For Manuel DeLanda, it is crucial that these agent-based behavioral models should be simulated using specific and singular individual agents, not abstract particles that embody the collective intelligence of an entire society (Leach 2009).
Figures 1 and 2_ Kokkugia: Roland Snooks & Robert Stuart-Smith Swarm Urbanism. Melbourne, Australia, 2009 http://www.kokkugia.com/swarm-urbanism
For the Italian Bernardo Secchi (2014), the science of urbanism cannot be easily classified. This is due to its interdisciplinary nature, which draws in equal proportion both on the past and on the future. In this sense, contemporary urban planners constantly work collaboratively with other disciplines, in processes involving both theoretical and practical experiences that take place over a period of time (Salazar Ferro and Ariza Parrado 2022).
An interesting proposal emerging from this perspective is the Kokkugia studio's 2009 urban redesign of the Melbourne Docklands3. The proposal's approach was to transform the existing urban network by expanding the business district. This was a speculative proposal that used emerging methodologies applied to the field of urban planning as a design tool and involved a sequential set of decisions at a reduced scale. It entailed the local interaction of independent agents without a sequential design hierarchy: instead of elaborating an urban plan, the designers programmed a conglomerate of self-organizing autonomous agents (micro or local resolutions), which interacted to generate a complex urban structure, leading to a system capable of responding in a flexible way to changing environmental conditions.
This was a simulation methodology, an algorithmic approach, based on swarm intelligence, which had the ability to generate programmatic relationships and architectural responses using a non-linear generative feedback process, and to develop an emergent self-organizing design using an adaptive logic procedure, just as an ant colony or a swarm of bees would in nature. The system used event-driven algorithms to obtain the desired results. Essentially, it was an innovative proposal involving an analytical/project management tool capable of predicting the possible evolution of a city in the immediate future. It had the capacity also, to develop subsequently into a state-of-the-art planning tool, with sufficient flexibility to adapt to the changing needs of the environment and respond to the stimuli, interactions and unpredictable acts of its inhabitants.
The model progresses from the city to the building. It is based on the behavior of elementary digital agents that interact in a predefined space and are calculated in real time, in a process involving billions of calculation steps. It is an ideal, collaborative prototype, with a graphical user interface that supports interaction, communication and collaboration between its various elements: an exploration of the benefits of using infrastructure-centric swarm systems that employ parametric algorithms for analytical optimization. It opens the door to infinite results.
For Roland Snooks and Robert Stuart-Smith (2009), Docklands is an insight into the emergent nature of public spaces, focused on the real-time behavior of digital social insects acting according to previously established rules. It is an optimized design that emerges from the behavior of each individual agent's neighbors and the traces they leave in its environment. It is a collective adaptation that is modified by external environmental forces operating in two stages of action. In the first stage, the agents self-organize the program through a process of stigmergic growth, a collective procedure whose logic is similar to that of termite colonies, which grow through the accretion of material to form mounds (the urban matter). In the second stage, the agents codify the elements and the urban topology, following a behavior similar to the self-organizing processes of mucilaginous molds, which produce systems with minimal pathways. In the Kokkugia studio project, the agents generate a collective intelligence to create infrastructure and circulation networks and complex three-dimensional systems. The process involves continuous feedback loops that encourage improvements and eliminate errors.
For its creators the challenge of this project did not lie in their ability to simulate current conditions, but rather to devise the operations and transformations involved in an emergent conception of cities: a shift in thinking that replaces master plans with master algorithms, a flexible process that synchronizes the micro and macro decisions produced during the urban design process: what Leach calls "devising operational processes and much higher levels of abstraction that involve seeding design intent into a set of autonomous design agents capable of self-organising [sic] into emergent urban forms" (2009, 56-63).
Figures 3 to 7_ Kokkugia: Roland Snooks & Robert Stuart-Smith Swarm Urbanism. Melbourne, Australia, 2009 http://www.kokkugia.com/swarm-urbanism.
Traditional urban planning tends to perceive the evolution of the city as a linear and predictable process, based on statistics and projections. However, the motors of urban transformation are complex and dynamic and face uncertain futures marked by fluctuations in the forces and energy flows that the city generates, shares and exchanges with its environment (Aquilué Junyent and Ruiz Sánchez 2021). Essentially, these fluctuations might be predictable, given that there are a finite number of options. However, despite this, it is not possible to predict which of these fluctuations will manifest and be amplified (Wagensberg 2003). Change is a non-linear process, meaning that as the system deviates from its equilibrium, its complexity -understood as the number of potential, accessible, solutions available to the system- increases (Aquilué Junyent and Ruiz Sánchez 2021).
Consequently, the development of advanced digital technology has allowed scientists and urban planners to study swarm intelligence in order to address the constant and unpredictable changes that occur in cities. Events such as natural disasters, fires or terrorist acts can be analyzed by algorithms based on the collective behavior of groups of agents, such as those used in swarm intelligence, making it possible for example to interpret the ways groups of people behave during an evacuation and to accurately determine the most efficient exit routes. These algorithms use metaheuristic methodologies to explore the behavior of the individual members of swarms, obtaining, as a result, optimal solutions to a variety of problems.
One of the most distinctive advantages of the method is its decentralized, or multi-agent, organization, with no centralized control or hierarchical logic. System processes therefore work to optimize individual and local objectives in order to attain the best possible overall performance. For this purpose, each component can be modelled as an autonomous agent, capable of making its own individual decisions. These agents collaborate within a multi-agent system to improve their performance and problem-solving capabilities at different scales (Herrera et al. 2021).
In the case of Melbourne, all elements of the urban fabric were conceived as autonomous agents, that is, as pieces that can interact and make decisions according to a hierarchy of intensities, and without a sequential design category. These agents behave as autonomous units; they have their own information sources and aspirations and are programmed to interact with each other within a self-organizing process that resembles that of natural selection described by Charles Darwin. Instead of conceiving the city as a compact mass, it is seen as a dendritic structure, based on the interplay of proximity and spatial principles (Batty 2009). The use of bio-inspired digital models in urban planning results in a rich and suggestive exploratory perspective with which to examine the behavior of cities: the processing of mass information, based on the use of algorithms, can lead us to predict and analyze problems before they occur.
A similar improved optimization technique is provided by the Grand Tour Algorithm (GTA), developed by Meirelles et al. (2020), which is based on swarm physics and uses a metaphor derived from the behavior of a peloton of cyclists, in order to calculate optimal solutions. Its attributes include resistance, defined by the distance to the leading cyclist (who represents the optimal solution) and speed, which is determined according to the difference between two consecutive evaluations of the target function. These two elements are then employed to determine the distance coefficients of each cyclist. In the end, the group will seek to follow the rider who is closest to the goal but will also follow the fastest rider at any given time. Consequently, the performance of GTA is superior to other classical algorithms, which, given its ease of use coupled with its speed of convergence, makes it a reliable solution, with a capacity to handle up to 20,000 decision variables, without the need to adjust its initial parameters to achieve a good optimization level (Meirelles et al. 2020).
Currently, the use of tools based on swarm intelligence in combination with artificial intelligence systems is transforming urban planning into a dynamic and variable system, capable of creating new architectural forms, returning value to historic centers, optimizing traffic routes, renewing green spaces and implementing new urban facilities. This is a process that is expected to lead to the development of more sustainable cities for future generations.
1 The acronym boids stands for bird-oid object, referring to objects resembling birds. Its pronunciation is similar to bird pronounced with a stereotypical New York (more precisely Brooklyn) accent.
2 However, by zooming in on the particles, a complex behavior can be observed in which all particles move in subtly different directions.
3 Founded in 2004, Kokkugi is a development and research platform founded by Jonathan Podborsek, Roland Snooks and Rob Stuart-Smith, three graduates of the Royal Melbourne Institute of Technology, which has campuses in London, New York and Melbourne.