Semiramis

Four core innovations

The project team on Gramazio and Kohler talks about four key innovations developed and used in the Semiramis track:

machine learning model:
The so-called “auto-encoder” – an artificial neural network developed in collaboration with the Swiss data center – was trained in a way that maps both the shape and spatial arrangement of the shells, including their performance in terms of sun and rain protection and arable area. According to the professors, this allows the design process to be reversed, because architects can now focus fully on design and explore the freedom of design.

Interactive design tool: The researchers understand that a second innovation is the interactive design tool, which was developed with the ETH Computational Robotics Laboratory and integrated into existing 3D modeling software. With the help of this software, the complex geometry of faceted casings can be easily modified while maintaining maximum values ​​such as inclination, volume and weight. At the same time, the structural bearing capacity is improved. If you move one point inside the geometry, the tool automatically adjusts the entire structure while respecting the specified parameters.

Collaborative robotics: For the complex multi-robot assembly process achieved at ETH’s Robotic Fabrication Lab, Gramazio Kohler Research has collaborated with Alphabet’s robotics software company, Alphabet. Intrinsic has developed a motion and path planning solution for robots that can be used with COMPAS FAB – an open source framework for digital fabrication, which was in turn developed at ETH.

Manufacturing and assembly: The project team sees the fourth innovation in manufacturing and assembly, especially with regard to items that can be disassembled for easier transportation. To do this, the five shells were divided into 14 pieces, which, in turn, consisted of 16 to 56 plates. These are linked to the use of a new type of joint sealing technology developed over decades of collaboration between Timbatec and the Berne University of Applied Sciences at Biel and ETH. This process enables a rigid connection to the edges of larger planks via butt joint jointing by injecting a special molding resin into the 3mm-wide joints between the panels placed by the robot.