Artificial Growth and Developmental Plasticity

Previous

Next

 

We have developed a new model of combined artificial evolution and growth in order to generate robots that exhibit interesting behaviours. Genomes encode genetic regulatory networks, which set of a set of a dynamical processes that transform a starting seed into a fully functioning, behaving robot. Our earlier models also included growth, but the growth programs of the robots were insensitive to environmental cues. Plants are an extreme, and instructive example of how nature combines genetic and environmental factors (in the case of plants, this includes light levels, hydrostatic pressure and soil chemistry) in order to produce a functioning organism. In our new model, environmental cues such as pressure and joint stress are transduced into regulatory chemicals, which can influence gene action. However, artificial evolution is not forced to produce genomes which respond to these chemicals: selection pressure may generate robots that grow based on a purely genetic developmental programme. However, in most of our initial experiments, selection pressure does produce genomes that exploit these signals, and use them to direct growth.

 

Figure 1: In the above figure, the first row shows a series of images from the growth of a robot evolved to grab an object in its environment (the large sphere) and lift it as high as possible in a fixed time period. This evolutionary run did not produce an agent that could lift the object, but it did evolve to rub its anterior appendage across it. The upper row shows the original, wild type robot. The lower row shows the same robot, but in this case it is not able to genetically respond to the environmental signals of touch and joint stress. Note the localized phenotypic differences: this indicates that this robot has evolved to grow specific morphological structures in response to environemental cues. Click here to see an animation of the original, wild type robot, and here to see an animation of the lesioned robot.

 

Previous

Next

bongard@ifi.unizh.ch