Very cool! My only gripe of low discrepancy sequences is how they are used in Quasi Monte-Carlo integration:

   1/n \sum_1^N f(a_n),

so each point gets equal weight. One can generalize this that every point gets it's own weight:

\sum_1^N w_{N,n} f(a_n).

I have never seen this generalization elsewhere. I only study Quasi Monte-Carlo as a hobby, maybe I just missed it. Once I had significantly improved discrepancies (close to theoretical minimum) with a modified Van der Corput sequence and linearly diminishing weights. It's all in one dimension though.

My other idea for two dimensions were to use the Hilbert-curve (the infinite limit) to map a one-dimensional sequence to a square area, but AFAIK this was done by others before.