Questions for SB3 Script about sumo-rl HOT 5 CLOSED

By the way-

I linked to this repo in the PettingZoo and Gym third party env lists:
https://www.pettingzoo.ml/third_party_envs
https://github.com/openai/gym/blob/master/docs/third_party_environments.md (this will be a website in the coming months)

from sumo-rl.

(new thread to simply the discussion of the hanging issues in #45)

"When using LIBSUMO, it is not possible to instantiate more than 1 simulation at the same time. This means that eval_callback can't work. If you use TRACI, I can easily implement multi-client support and this would be possible. But remember that TRACI is way slower than LIBSUMO, so I'm not sure whether this is advantageous." -Could you please elaborate a bit more on all this? Using eval callback is essential to what I'm planning to do with this, and running multiple environments at once is generally desirable when learning with PPO

I have just added support for multiple simulations in parallel with TRACI :)
TRACI and LIBSUMO are different libraries (with basically the same API) that you can use to control SUMO through python. TRACI is way slower, but lets you open multiple parallels connections (and thus instantiate training and eval envs at the same time). LIBSUMO is way faster, but for now it only supports a single connection.
In short, if you need to run multiple envs in parallel, you need to use TRACI. Then, unset the LIBSUMO_AS_TRACI system variable.

Regarding how rendering doesn't work in my sb3 code snippet, is there code somewhere that generates videos of the simulation rendering that I could adapt into that snippet?

Give a check at https://sumo.dlr.de/docs/FAQ.html#building_videos_from_sumo-gui . I have never done this, but it should be possible.

Thanks again for all your help, I really appreciate it :)

I'm glad to help! :)

from sumo-rl.

By the way-

I linked to this repo in the PettingZoo and Gym third party env lists: https://www.pettingzoo.ml/third_party_envs https://github.com/openai/gym/blob/master/docs/third_party_environments.md (this will be a website in the coming months)

Thanks! :D

from sumo-rl.

Do you happen to have any benchmarks of how much slower TRACI is vs LIBSUMO?

from sumo-rl.

Do you happen to have any benchmarks of how much slower TRACI is vs LIBSUMO?

I have tested on experiments/ql_4x4grid.py once, and LIBSUMO was around 8x faster. However, the bottleneck is the number of vehicles, as we need to retrieve information for all vehicles in the simulation. So in smaller scenarios this difference may be not that big.

You can try to ask in https://github.com/eclipse/sumo if they have more results on this.

from sumo-rl.