[Enhance] Real scenario adjustment - Githubissues

LovelyBuggies / sumo-gym

OpenAI-gym like toolkit for developing and comparing reinforcement learning algorithms on SUMO.

Other

30 stars 7 forks source link

[Enhance] Real scenario adjustment #2

Closed LovelyBuggies closed 2 years ago

LovelyBuggies commented 2 years ago

Some implementation doesn't match the real traffic scenario well:

Vehicles don't need to avoid cash (though VRP does).

https://github.com/LovelyBuggies/sumo-gym/blob/18aa7ca92cc21e945066f9a2c9b3f01567273e8b/src/sumo_gym/spaces/network.py#L43-L48

When sampling, to get more rewards, we need to bypass to get more profits. But currently, we pass in a fully-connected graph to overcome this issue. Probably need a search algorithm to find the minimal-cost route to finishing the task.

https://github.com/LovelyBuggies/sumo-gym/blob/18aa7ca92cc21e945066f9a2c9b3f01567273e8b/src/sumo_gym/spaces/network.py#L33-L38.

LovelyBuggies commented 2 years ago

finished.