I'm working with Vytas at NASA on the control issues for these tensegrity robots. One of our main ways to address the complexity issue is to use machine learnin

I'm working with Vytas at NASA on the control issues for these tensegrity robots. One of our main ways to address the complexity issue is to use machine learning to determine control parameters. We have been using traditional evolutionary techniques as well as some techniques related to importance sampling. A counterintuitive result of using evolutionary controls is that the control problem does not necessarily become more difficult as the complexity of the structure increases. This is because the number of good solutions may increase faster than the solution space. Currently in hardware, we are building relatively simple structures, but we hope to expand to more complex ones in the future.