Видео

It's where the wild things are.

Hi, Thanks for your question! Only Anya searches over a grid. Polyanya searches over a navigation mesh. If you have a grid-based representation, it's easy to construct a mesh and doing so usually gives substantial performance benefits. Conventional methods for searching navigation meshes place graph nodes in the middle of polygon faces, or along polygon edges. These graphs are small, fast to update, and fast to search but they do not produce Euclidean-optimal solutions; it's easy infact to construct examples with substantial detours. But is this method faster than Polyanya? That probably depends on the implementation, but I expect similar performance in general. This is because Polyanya expands few intervals from along a given polygon edge, similar to nav-mesh approaches. It may even be faster because we perform aggressive pruning. We never ran the experiments because the comparison is apples-to-oranges (optimal vs. non-optimal) and because the baseline (Anya) was already shown to be competitive with preprocessing-based suboptimal solvers. Visibility graphs are another type of navigation graph. These require substantial offline preprocessing but they guarantee solutions which are Euclidean-optimal. Once constructed, visibility graphs are usually fast to search. I say usually because again it's easy to construct examples where visibility graphs are very dense, which is bad for search performance. Under favourable conditions, I think visibility graphs might be competitive, in terms of speed, with Polyanya. But they have so many disadvantages which makes this approach difficult to recommend (expensive preprocessing, large storage costs, relatively expensive start-target insertions, branching factor explosion).

Hi @BernardoTelles. Thank you for reporting the issue and thank you for your kind comments regarding the MAPF Tracker! We can confirm there were some interruptions in the last few days which caused the system to be unavailable. These issues have been fixed and everything should be working once again. We do have a github repo (github.com/ShortestPathLab/MAPF-benchmark-web) with source code for the system but the repo does not contain any of the uploaded results (the dataset is very large and unwieldy). The good news is that all of the data can be downloaded (in a compressed format) from our website. You might find the github repo useful to report issues or request features.