B-Pro Design Computation Lab - Research Cluster 4

PizzaBots aims to investigate autonomous processes in design and construction, with a focus on the housing sector. By designing an identical geometric syntax between robotic process (active pieces) and building components (passive pieces), the project is able to transform an end-effector into an active component of the design process using Machine Learning.

Using a modular elements and a finite number of iterations, the project achieves a more efficient workflow and minimises resources used. A 5x5 grid system is analysed to find combinations of individual tiles that can produce an optimum amount of possibilities. After a first research design, a small box shelter generated using pathfinding algorithms, the team utilise the logic derived from movement and assembly sequence of the robots to arrive at their assembling logic. This results in two main types of connection–Straight (straight robot movement) and Corner (rotational robot movement)–which cover all possible ways to join two tiles together.

In order to further automate the process, PizzaBots utilises a multi-layered neural network with a reinforcement learning model to train the robot to pick tiles and place them into predefined locations, and repeat this process. Machine Learning is also applied to address the issue of potentially uneven terrain, and a counterbalancing strategy enables assembly of the structure without scaffolding.

The autonomous system recalculates the climbable surface for the robot everytime a new voxel (building block) is generated, and is supported by finite element analysis to assign direction of growth in line with structural behaviour.

The user decides the starting point of the structure based on basic program requirements (live, eat, cook, sleep, etc). Then the robots are deployed. Because the objective of the research is to provide low-cost houses and keep the structure light-weight, the main building materials are OSB Boards, glass and aluminium. The individual boxes are joined using metal slider panels.

The system allows for on-site manufacturing, and an automated assembly process. Every built structure requires a ratio of 1% robots to number of building components (tiles, glass and aluminium boxes) required for completion. The tiles can be fitted with electrical, water and heating infrastructure, making the aggregated structure a fully inhabitable space.

Assembler Assemble - PizzaBots

B-Pro Research Cluster 4, the Bartlett School of Architecture, UCL

Design Tutors: Gilles Retsin, Manuel Jimenez Garcia, Vicente Soler

Theory Tutor: Mollie Claypool

Students: Mengyu Huang, Dafni Katrakalidi, Martha Masli, Man Nguyen, Wenji Zhang