search

siyandong / NeuralCoMapping

[CVPR 2022] Multi-Robot Active Mapping via Neural Bipartite Graph Matching.
MIT License
82 stars 7 forks source link

About Sinkhorn output #9

Closed JiayunjieJYJ closed 1 year ago

JiayunjieJYJ commented 1 year ago

Thank you for your excellent work, and I have an issue I'd like to discuss with you. Sinkhorn's output is the result of several rounds of raw and column normalization, but I wonder what happens if two robots pick the same frontier? For example, the following Sinkhorn's output ： image final matching matrix: image

How does your code solve this problem?

Thanks!

illusive-chase commented 1 year ago

It would be possible that two robots pick the same frontier (they will just try to move towards the same goal position). However, a well-trained policy will avoid that to achieve higher efficiency.

JiayunjieJYJ commented 1 year ago

It would be possible that two robots pick the same frontier (they will just try to move towards the same goal position). However, a well-trained policy will avoid that to achieve higher efficiency.

OK, thanks! Now I have more problems reproducing it. Hope you can help me. Thanks!