During the ‘translations’ course with Alper Semih Alkan, at TU Delft, I studied two competition entries for Parc de la Villette and noticed that both projects attempted to create not a specific park for a particular location, but rather to generate a new way of designing a park - almost, like it would be a recipe for a park.
And since recipes can be replicated, I wrote a script that randomly generates my version of Parc de la Villette each time I press F5.
Original designs:
The design of the park de la Villette is one of the flagship symbols of Paris. The winning concept by Bernard Tschumi is a reflection of what an “urban park for the twenty-first century” should look like. Breaking with the conventional understanding of the park, the author uses the superimposition of three means of expression: point grid, lines, and surfaces, to arrange the complex program using programmatic measures. The project is an innovative way of thinking about both space organization and architectural representation.
The other key project is the competition entry by Rem Koolhaas and OMA. Searching for new methods of organization, the project uses the superimposition of strips, point grids, lines, and large-scale objects as a strategic assessment of programmatic concepts. The project presents a new, flexible architectural typology, supported by mathematical distribution. “The proposed project is not for a definitive park, but for a method that - combining programmatic instability with architectural specificity - will eventually generate a park.”


The script:
A very crude yet functional Grasshopper script. I've highlighted in red the components responsible for adding "randomness" to each ingredient in the recipe.




The output: 100 versions of de la Villette by Tschumi

Effectiveness of the script.
A manual process tests the completeness of the output, balance and line arrangement of the generated images, as well as the collision avoidance between surfaces and a point grid.





The output: 100 versions of de la Villette by OMA

What is the quality of the generated output?
I've looked into the patterns of similarities between produced images. Starting with a matrix of 100 generated propositions, the process of manual filtering searched for the relations between them, such as overlapping, separation, incompleteness, parallelism, and line intersections.





I guess the main takeaway from that was that what might be easy to describe hides many complexities we are not aware of.
The attempts to recreate design processes using scripts allow us to better understand the complexity of used tools and the relations between them. Various subconscious processes remain invisible until we materialize them.
Thanks for reading