Exterior View of Built Form
It broadens the horizons of the research-based approach from extraordinary design synthesis to robotic manufacturing processes and encourages a dialogue between them.
The complex design strategy is tested on three different scales:
1. Small fabrication chunks.
2. Habitable architecture chunk prototypes.
3. Built form for NGV Contemporary.
1. Small fabrication chunks.
2. Habitable architecture chunk prototypes.
3. Built form for NGV Contemporary.
Project Plan
It deals with bottom-up adaptive self-organizing algorithms to achieve complex forms, robotic manufacturing processes especially spatial and spray concrete 3D printing, and machine learning frameworks such as Reinforcement Learning.
The project involves a framework / genotype with goal-oriented agent behaviors and tectonics, mathematics of voxels and reinforcement learning frameworks. It involves a differential growth algorithm for form finding, voxelising generated forms and further replacing them with some geometries using Reinforcement learning.
It follows a “bottom-up” approach where the system grows and evolves up to the event of an emergent and self-organized behavior, obeying the simple rule about the interactions of an agent-based system. There are internal and external parameters affecting the system related to local conditions and environmental and urban planning data.
Exterior View of Built Form
The built form is adaptable to the underlying conditions and also in terms of allowing continuous gradients of structure, aperture, porosity and surface articulation. The overall built mass portrays coherence and differentiation with a solid base and a complex porous form levitating over it.
Architecture Chunk Prototypes
Architecture Chunk Prototypes
The complex form of the chunks is like a continuous single mesh which ripples into a slab, wall, or a staircase thereby decluttering the space and creating such characters.
RL Algorithm Actions
The number voxels or discrete units is in the range of millions or billions it is difficult for us as humans to perform the calculation of which component geometry and in what orientation should replace a particular voxel.
RL Algorithm Rewards System
To replace the voxels with components, the aim is to select the components for each voxel while maximizing the rewards for a set of established criteria such as Density, Orientation, Structural Stability for the whole aggregation, Toolpath Connectivity for 3D printing etc.
Reinforcement Learning Framework
To solve the issue, the proposal seeks to envision an environment where a Machine Learning paradigm evaluates and optimizes the configurations. It incorporates the use of Reinforcement Learning where an agent is trained to take actions in an environment to maximize rewards.
RL Algorithm Simulation
The algorithm works through trial-and-error testing multiple component possibilities for each voxel until it converges to the most optimal configuration. It dwells on the notion of computational design and explores the idea of cognition in discrete architecture. It involves a bottom-up approach where discrete elements aggregate to generate larger spatial configurations while catering to multiple design goals within part to part (Local) relations and part to whole (Global) relations or ‘mereology’.
The component assemblies are then evaluated and analyzed by AI algorithms trained to optimize the part-to-part connections to ensure more intuitive top – down decision making and enhance the performance of the large assemblies.
3D Printable Chunks
The algorithm segregates the form into these small chunks which can easily be 3D printed and attached together afterwards to create the larger forms.
Local Deformation
Cohesion Deformation
Deformation of the component geometries is done based on the idea of topological deformation by D’Arcy Thompson. It helps break the order and regularity of the underlying discrete square grid. 3D printing as a strategy provides the opportunity of fabricating variations of a toolpath instead of printing the same toolpath repeatedly by retaining the discrete properties of the underlying curve path.
Architecture Chunk Prototypes
The following chunk trials depict how the deformation affects the overall form breaking the linearity of the grid creating something more organic and continuous.
Fabrication Chunk Prototypes
The fabricated prototypes are the first scale for testing the algorithmic processes and fabrication strategies. These forms with bifurcating and converging strands blur the distinction between solid and void and structure and ornament.
Architecture Chunk Prototypes
Top View of Built Form
The variation of density in the component geometries leads to a juxtaposition of characters of solid and void, thereby creating an effect blurring the transition between different characters of design. As visible, the character goes form completely solid areas to completely porous and light areas.
Architecture Chunk Prototypes
Multiple iterations of the chunk prototypes were tested to achieve varying characters of deformation, porosity, intricacy, directionality, and patterning.
Architecture Section
The overall form and articulation is driven through the interaction of different algorithmic processes, material etc. The geometry negotiates complex behaviors such as structure and ornament, generating emergent characteristics that shift throughout the project.
Interior View of Built Form
The different characters of the architecture elements and how they merge with each other is visible in these views. The gradients of porosity and aperture in the form leads to an intricate fabric of architecture and a high-resolution fabric of light.
Interior View of Built Form
The idea tries to bring both the human designer and the machine (Artificial Intelligence) in a symbiotic relationship. The intent of the designer is partly captured in the generative algorithm as behaviors and partly in evaluation encoded in the designer’s intuition. The AI algorithm acts as an extended mind to the designer and helps to design and realize such high-res heterogenous intricate forms.
Through a constant interaction between the top -down and bottom -up strategies as depicted, the proposal seeks to produce an adaptive and responsive architecture.