The quantspace package implements a version of quantile regression developed by Larry Schmidt and Yinchu Zhu in this paper. An issue with quantile regression is that quantiles of the predictive distribution are always ordered monotonically, but any quantile regression will have quantiles that eventually cross. Rather than working with the level of the predictive distribution, they translate this to a regression on the log difference between quantiles, which prevents them from crossing anywhere in the predictive distribution. You can find the documentation here, although development is dormant.