Our project starts in a web browser where a user can define the input parameters of a tree-like bookshelf. Based on the trained AI data, the user will also be able to see the structural performance of the chosen bookshelf, which will not be a proper structural analysis, but a hint/ indication for good or bad structural performance.


We have tackled the L-systems structural displacement because tree structures are notoriously difficult to predict in terms of structural performance and often give surprising results. 

Throughout the development of the project we have reviewed a number of generative options to create an L-system – we have used Rabbit and Anemone, but since we required a very high level of control of the system and data export, we developed our own L-system script, which is not based on a plug-in. The user can control everything related to the shape – the trunk height, number of branches, heights of branches, etc. In the future development of this project, we see even further opportunities to link the generative parameters to Karamba. For example, there could be a list of materials and profile thicknesses that the user would be able to pick as well as shelves shapes and sizes. For the purposes of this course, we have limited the input parameters from the user to 7 and the features to 17. And based on this we exported the final dataset.

ML results and tests 

We have decided to use a regression ANN model, a min-max scaler for both the input (x) and output (y) as these were giving better results, the loss function as mean squared error and the optimizer as Adam.

When it comes to visualizing the results, there is a limitation due to Karamba functionality, you can’t feed displacement values to generate geometry. For this reason, we have used a simple numerical comparison between actual displacement and predicted displacement. In the future of the project, we would compare the results by training the dataset from B to A. In general, so far the ML model is giving us promising results for the numerical displacement predictions.

Tree Bookshelf is a project of IAAC, Institute for Advanced Architecture of Catalonia developed in the  Master in Advanced Computation for Architecture & Design  2020/21 by Amar Gurung & Polina Hadjimitova  and Faculty: Gabriella Rossi and Iliana Papadopoulou