Multi-agent system in architecture

1.Introduction

In today’s architecture, digital design is still dominated by Spline modellers. Mario Carpo (2014) discussed that designers use spline modellers “model” reality through converting it into a mathematical script, and the continuous lines and uniform surfaces they draw or make are entirely only a material approximation of the mathematical functions that computers have calculated for them. As indeterminacy and nonlinearity are now a fact of our daily life, embedded in most digital technologies we use. But splines are mathematical objects and a tool of simplification; they do not fit with the phenomenological world we inhabit. (M Carpo,2014).

Multi-Agent Systems(MAS) are one computational territory where generative design opens toward complexity. Specifically speaking, MAS are introduced to solve complicated, real-time problems in an uncertain and changing environment. (N Glaser,2002) With MAS, it is possible for designers to explore non-liner, emergent, and behaviour modelling in architecture.

This paper is a review of background knowledge and some main features of MAS. The primary objective is to give a clear understanding of the topic and how MAS help to reveal a resilient and complex new fabric of architecture.Add paragraph text here.

2. From Agent to MAS

The term Agent comes from the Latin word “agentem”, which means “to set in motion, drive, lead, conduct”. In computer science, an agent is best defined as “a component of software and/or hardware which is capable of acting exactingly in order to accomplish tasks on behalf of its user.” (Hyacinth S. Nwana,1996). According to N Gilbert (2008), agents are described as ”either separate computer programs or, more commonly, distinct parts of a program that are used to represent social actors—individual people, organisations such as firms, or bodies such as nation-states.” They are devised to react to the computational environment which is a model of the real environment in which the social actors operate. (N Gilbert,2008).

Regularly, agents are simplified entities that achieve direct and simple goals. Multiple agents can be assembled in swarms which operate collectively to achieve larger goals, a concept described as swarm intelligence which will be discussed later (G Beni and J Wang,1989). MAS are an increasingly important way to solve complex problems or model complex systems in areas ranging from ecology, economics or cities (Olfati-Saber, Fax, and Murray, 2007). Take cities as an example, in the work of Manuel DeLanda, he applies MAS to urban simulation, as he mentioned that agency must be attributed to concrete, singular, individual entities: persons, communities(when they form coalitions in social justice movements) and institutional organisations. Furthermore, Manuel DeLanda places great emphasis on decision-making agents. After intelligent decision-making agents can be programmed, we would be in a position to simulate the growth of real cities (N Leach,2009).

3.Flocking Behaviour and MAS

According to A Andrasek(2012), DJ Gerber, E Pantazis and LS Marcolino(2015), MAS in generative architecture are mostly based on the Craig Reynold’s algorithms for biods (Figure 1, Figure 2, Figure 3) which are examples of Swarm distributed behaviour. These forms are the result of the continuous multi-port data system, so they come with features of various types of input data, showing the unique transition into organic natural forms. This design system is flexible; they always show a mixed, fixed, occurrence and disappearance, dispersion and agglomeration of macroscopic effects.

As Craig Reynold (1987) discussed, the simulation of a flock of birds is based on modelling the behaviour of each bird independently by regarding it as an agent. This approach assumes that a flock is simply the result of the interaction between the behaviours of individual birds. The birds try to stick together and avoid collisions with one another and with other obstacles in the environment (CW Reynolds,1987).

When it comes to MAS in architecture, agents are programmed through the features of cohesion, alignment and separation (Figure 4) and variable values of their spring connections. (A Andrasek, 2012)

Following are the exploration the author played with the three forces. The result is achieved in Processing computation platform.

4. Key Concepts of MAS

Flocking behaviour discussed above, as well as ant colonies, are examples in natural systems of Swarm Intelligence(SI). E Bonabeau, M Dorigo and G Theraulaz (1999) extended the definition of Swarm intelligence to include any tempt to design problem-solving devices inspired by the self-organized behaviour of social insect colonies and other animal societies. According to A Andrasek(2012), TD Wolf and T Holvoet (2005), E Bonabeau, M Dorigo and G Theraulaz (1999), the following are some of the key concepts about the MAS.

­Simplicity to complexity. Simplicity is required in the design of any system in nature. It is possible to explain the complex collective behaviour of insects by assuming all individuals are relatively simple interactive entities (E Bonabeau, M Dorigo and G Theraulaz,1999). There is not a central control in the swarm phenomenon. Individuals who need not be identical just follow simple rules, it is their interactions that affect the system. When agents interact collectively, they are capable of producing complex behaviours and emergent effects (A Andrasek, 2012). For instance, the beautiful complex behaviour of flocking birds can be stimulated in three principles:1.Collision Avoidance: avoid collisions with nearby flockmates. 2. Velocity Matching: attempt to match velocity with nearby flockmates 3. Flock Centering: attempt to stay close to nearby flockmates (CW Reynolds,1987).

Scalability and Autonomy. MAS are used to model self-organising systems. The system gets more organised by the cooperation and group formation in MAS. Firstly, scalability here refers to” Increase in Order”. With the number of agents increasing, the system has more strength. “So as to promote a specific function” a system should not be too much order or no order. It needs to find a balance so that can exhibit a flexible and organised behaviour. Secondly, without autonomy, a system cannot be regarded as self-organized. In other words, it needs to organise without interference from the outside (TD Wolf and T Holvoet,2005).

Robustness and flexibility. No single agent can represent the global emergent in self-organizing systems. There is a possibility that the system can stick to its function even though some individuals may fail to perform their task. This flexibility allows for that individual entity to be replaced (TD Wolf and T Holvoet,2005). Take the flocking birds above as an example, birds in a flock can be substituted by others, yet the flock phenomena remain. That is to say, a self-organizing system can absorb vagueness and noise which is not contained in binary logic compared with triadic logic.

Randomness and Openness. The behaviour of the individual agent in swarm intelligence is often a balance between a simple perception-reaction model and a random model. Although it could cause ineffectiveness in the system, randomness is very crucial since it enables that a better solution will be found. No solution is the final one. Once with a positive feedback mechanism, the new configuration will form. Therefore, the randomness can characterise the agent’s behaviour with a certain degree of openness (E Bonabeau, M Dorigo and G Theraulaz,1999).

Adaptability. Single agent solution may not be enough since the complexity of the design problems. A system with multiple agents, each one responsible for a different aspect of design is necessary. For example, some agents can absorb large data related to the environment, while others responsible for creating the form. When the system exhibits a large variety of behaviours, adaptability is crucial. Adaptability can be formulated as:” a change in the environment may influence the same system to generate a different task, without any change in the behavioural characteristics of its constituents”. (T D Wolf and T Holvoet,2005) This concept is similar to “stigmergy” of swarm intelligence. A system with high level of adaptability is priceless: agents can respond to a perturbation without being reprogrammed. (E Bonabeau, M Dorigo and G Theraulaz,1999).Add paragraph text here.

5. Data Architecture and MAS

Architecture is regarded as a form of artificial life (J Frazer,1995) and has drawn inspiration from nature ---from its forms, structures and the inner logic of its morphological processes. Natural ecosystems have a complex biological structure that most man-made environments do not have: they recycle their materials, permit change and adaptation, and make efficient use of ambient energy. (J Frazer,1995). Von Neuman, a key figure in the development of computing, pointed out that the basis of life was information encompassing both natural and artificial biology.

Nowadays, the emerging reality is that the continuity of space is challenged by a global flow of data. Using the resources of extreme computing, it enables architecture to embrace the data-driven existence (A Andrasek and D Andreen,2016). As the architecture can be regarded as a container for energy production and its fibres can reach higher levels of performance, it is possible to descend into the cellular grain of matter, flow of light, heat, vapour and friction ( A Andrasek,2012). Furthermore, “there is a possibility to crack the behavioural tendencies of matter at a finer grain. In such molecular compositions, every particle counts.” ( A Andrasek,2012). In this case, MAS have inherent ability to deal with the large data as Andrasek(2012) discussed in her article: “A collective entity designed out of such a complex choreography of neighborhood-based computing acts as an interface for a non-linear negotiation between multiple data sets, etc.”

Following are two works from Research Cluster 1 at the Bartlett Graduate Architectural Design programme. They are distributed multi-agent-based systems coded through object oriented programming and physics simulations, often in combination with high accuracy robotic fabrication methods. (A Andrasek and D Andreen,2016).

a.Fluid motion (Figure 5), in this project, a simulation of impossible liquids form a topological landscape where agents arrange matter in patterns suitable for robotic polymer weaving. The simulated physics makes it possible for designers to generate complex behavior by manipulation of viscosity, surface tension, mass/gravity, magnetism and similar physical properties. Through the variable properties, physics field is able to adapt to his host environment (A Andrasek and D Andreen,2016).

b.Crystal Cloud (Figure 6), in this project, thousands of agents rearrange a lattice structure.”The agents interact locally and continuously scan the structures they meet, searching for conditions that trigger theminto action: analyzing the local neighbourhoods, tracing the force lines that support the structure and amend vertical discontinuities, removing redundant particles, measuring the penetration and diffraction of light beams as they pass through the crystal lattice and manipulate it through aggregation or disaggregation.” (A Andrasek and D Andreen,2016).

5. Design and Designer

In traditional architectural design practice, the designer is the creator of the form. The architects work through representative approaches, creating a model of the building, analysing and testing it mathematically and passing it on linearly to other actors like client and builders. When the process and object become intervened through mutual feedback from contextual data, fabrication and inhabitation, the role of the designer changes to that of a programmer, who sets behavioural rules and relationships. As a result, the attention of the architect is shifted from the visible, formal representation, to extend it into a domain of the invisible: the underlying logic and procedures.(A Andrasek and D Andreen,2016).

It is the increased volume of computational power that makes this design approach possible. This allows the designers to proceed from systems containing hundreds of parametric elements to the acute ones with millions of agents or billions of particles.

6. Conclusions

The literature review in this paper has concentrated mostly on some key concepts about MAS. It allows designers or modellers to model reality in a natural way, to represent the emergence of structure and various kinds of adaptation, none of which is easy to achieve by other approaches.

As agents intricately interlace with one another, Designers using MAS can reach geometric complexity by keeping individual agent simple. Regarding flexibility, MAS that can absorb vagueness and noise enable the operation of fertile tensions between difference and invariance. With autonomy, randomness and adaptability, agents can act autonomously, perceive and adapt to change, and create and follow goals.

When it comes to architecture, MAS with the characteristics discussed above provides a possibility for architects to find some free solutions of complex design problems and embrace the big data. As for the designer, their role will be entirely different from the past when adopting MAS in architectural design.

REFERENCE

1. CW Reynolds, Flocks, Herds, and Schools: a Distributed Behavioral Model. Computer Graphics, 21(4),1987，pp 25-34.
2. G Beni, J Wang, Swarm Intelligence in Cellular Robotic Systems, Proceed. NATO Advanced Workshop on Robots and Biological Systems, Tuscany, Italy, June 26–30 (1989)
3. DJ Gerber, E Pantazis and LS Marcolino, Design Agency, Prototyping Multi-Agent Systems in Architecture. Conference: Proceedings of ‘Next City’ CAADFutures Bi-Annual Conference, At Sao Paulo, Brazil, 2015.
4. TD Wolf and T Holvoet, Emergence Versus Self-Organisation: Different Concepts but Promising WhenCombined. ESOA 2004, LNCS 3464, S Brueckner, et al, eds. Springer-Verlag: Heidelberg,2005, pp 1-15.
5. A Andrasek, Open Synthesis //Toward a Resilient Fabric of Architecture. Log, 25, 2012, pp 45-54.
6. A Andrasek, D Andreen, Activating the invisible: data processing and parallel computing in architectural design, Intelligent Buildings International, 8:2, 106-117, DOI:10.1080/17508975.2014.987641, 2016
7. E Bonabeau, M Dorigo and G Theraulaz, Swarm Intelligence: From Natural to Artificial Systems, Oxford University Press: New York,1999, pp 1-24.
8. J Fraser, An Evolutionary Architecture, Architectural Association n: London, 1995, pp 9-21.
9. M Carpo, Breaking the Curve, Artforum International, R Morris and K Attia, eds, February 2014, pp 169-173.
10. N Leach, The Limits of Urban Simulation: An Interview with Mauel DeLanda, Architectural Design 79(4):50 - 55 · July 2009
11. N Glaser, Conceptual Modelling of Multi-Agent Systems, The CoMoMAS Engineering Environment. Kluwer Academic Publishers: Dordrecht, 2002, pp 1-12.
12. N Gilbert, Aged-based models. Guildford, UK: SAGE,2008.
13. Olfati-Saber, R., J. A. Fax, and R. M. Murray. 2007. “Consensus and Cooperation in Networked Multi-agent Systems.” Proceedings of the IEEE 95 (1): 215–233.
14. Hyacinth S. Nwana, Software Agents: An Overview. Knowledge Engineering Review. Vol. 11, No. 3, pp.205-244. October/November 1996. Cambridge University Press, 1996.