Isn tile-coding just grids, and thus subject to the curse of dimensionality?
Basically, no. Tile coding is a quite general idea and many of the ways it can be used avoid the curse of dimensionality. There are at least three common tricks: 1) you can consider subsets of the state variables in separate tilings, 2) you can use overlapping tilings to get vastly higher resolution in high dimensional spaces than would be possible with a simple grid using the same amount of memory, and 3) you can hash down to a smaller space so as only to devote memory to the portions of state space that actually occur.
