The initial aim of this project was to generate real modular robot morphologies by means of a cell-based developmental model. In this model, a cell was representing a module of the robot and was controlled by an artificial gene regulatory network, a simplified computational model of biological gene regulatory networks. In order to evaluate the complexity and the possible properties that could emerge with this approach, we have generated images with our gene regulatory network model. The obtained images were surprisingly attractive (see images in the supplementary material) by both their complexity and the artistic qualities. We were also able to point out interesting properties such as regularities (repeating pattern with slight modifications), symmetries and fractal-like properties (possibility to zoom in color transition that generates indefinitely increasing complexity).
Jordan Pollack and I have extended this work with an artistic consideration. We now propose an interactive tool to evolve gene regulatory networks and thus the images they generates. Moreover, the previous work was only taking into consideration the spatial aspect of the gene regulatory network to generate only images. We also have extended the capacities of our approach by using the temporal dynamics of the regulation in order to generate movies. The movies are evolved just like the pictures are.
The evolution consists of an interactive genetic algorithm, inspired by Richard Dawkins’ blind watchmaker and often used as an evolutionary art tool. The evolution is based on crossing and mutating a genome representation of regulatory networks with the aim to slightly modifying their structures and combine their properties. A special care has been taken to the design of the graphic user interface: few parameters are presented and an evolution tree is provided to the user to allow him/her to navigate in the evolution history.
Once a gene regulatory network is obtained, the corresponding image or movie can also be extended to generate a high definition media: one property of this approach is to be very scalable and thus very flexible to the size of the desired media.
Images :
Videos :
Tool for the interactive evolution of gene regulatory networks : GRN_Evolver.jar (92ko)
Tool to generate HD images and videos from evolved regulatory networks: GRN_Extender.jar (93ko)
Sylvain Cussat-Blanc (University of Toulouse - IRIT)
Jordan Pollack (Brandeis University - Demo lab)