Resilient Structures Through Machine Learning And Evolution (2013)
article⁄Resilient Structures Through Machine Learning And Evolution (2013)
abstract⁄In the context of the growing usefulness of computation within architecture, structures face the potential for being conceived of as intelligent entities capable of resilient, adaptive behavior.Building on this idea, this work explores the use of machine learning for structures that may learn to autonomously ‘stand up’. The hypothesis is that a neural network with genetically optimized weights would be capable of teaching lightweight, flexible, and unanchored structures to selfrectify after falling, through their interactions with their environment. The experiment devises a physical and a simulated prototype. The machinelearning algorithm is implemented on the virtual model in a threedimensional physics environment, and a solution emerges after a number of tests. The learned behavior is transferred to the physical prototype to test its performance in reality. This method succeeds in allowing the physical prototype to stand up. The findings of this process may have useful implications for developing embodied dynamic structures that are enabled with adaptive behavior.
|
|
| Year |
2013 |
| Authors |
Mehanna, Ryan. |
| Issue |
ACADIA 13: Adaptive Architecture |
| Pages |
319-326 |
| Library link |
N/A |
| Entry filename |
resilient-structures-through-machine-learning-evolution |
