Researchers have developed learning algorithms that let quadrotors perfectly balance ungainly poles — and juggle them around between themselves.