suboptimal paths with allow_diagonal=True about pyastar2d HOT 3 CLOSED

Actually, just after opening this issue I had the deciding insight: it seems the weights have to be np.float32, not np.float. So feel free to close this again.

On a sidenote: it would be nice to be able to specify a higher cost for moving diagonally as I am getting a lot of diagonal movements where it would be possible to just go straight, as e.g. here:

from pyastar2d.

it seems the weights have to be np.float32, not np.float

Correct -- this is the datatype expected by the underlying C++ implementation. Maybe there should be an explicit check of the datatype enforced.

it would be nice to be able to specify a higher cost for moving diagonally as I am getting a lot of diagonal movements where it would be possible to just go straight, as e.g. here

This implementation defines the shortest path as the path that minimizes the sum of pixels traversed, so making diagonal moves more expensive doesn't really fit with that. However, if you're working with grids with large areas of uniform-cost but prefer straight paths over diagonal ones it should be a fairly straightforward change to give precedence to straight paths by changing the order in which neighbors are added to the priority queue:
https://github.com/hjweide/pyastar/blob/master/src/cpp/astar.cpp#L92

But if you're working with uniform-cost grids, why not use jump point search instead? :)

from pyastar2d.

Thanks for your reply. An automatic check or a note in the readme might indeed be helpful :)
Makes sense in the pixel perspective. I will look at what you've mentioned.

from pyastar2d.