I am finding the MongoDB aggregation framework to be extremely powerful - it seems like a good option to flatten out an object. My schema uses a an array of sub objects in a
There is a request for something like this in jira https://jira.mongodb.org/browse/SERVER-5947 - vote it up if you would like to have this feature.
Meanwhile, there is a work-around if you know up front what the possible values of the keys will be (i.e. all unique values of "category") and I have some sample code on it on my blog.