I assume that future AGI which is smart enough to control "general purpose robotics" is probably smart enough to design robotic forms that make all the current

I assume that future AGI which is smart enough to control "general purpose robotics" is probably smart enough to design robotic forms that make all the current stuff obsolete anyway. Like the article says, physical world data is too scarce to jump straight to powerful robotics first.