seq2seq for pathfinding

Can we use these seq2seq matchers for robot planning and pathfinding? I.e. given points A and B, find point C, laying between them. Because dijkstra's is fucking slow and A-star is useless, when space in non-linear (i.e. teleporters are present).

>>1
Pro game programmer here, recently worked on an RTS for a big publisher. One of the things I did was optimize the path finding. We used Jump Point Search Plus (JPS+), which gave us an improvement of 2 magnitudes in performance over A* for our needs.

However, as of a three months ago, a Nintendo programmer iterated on JPS+ and came up with Jump Point Search Plus with Goal Bounding, which can be a few or even several magnitudes faster than A*, depending on how much data you have--it scales way better.

https://github.com/SteveRabin/JPSPlusWithGoalBounding