search

Guojyjy / CoTV

Cooperative control for traffic light signals and connected autonomous vehicles using deep reinforcement learning
MIT License
57 stars 11 forks source link

Different results #5

Closed TrinhTuanHung2021 closed 5 months ago

TrinhTuanHung2021 commented 10 months ago

Hello Thank you for uploading your models. I have implemented your models with CoTV under the Dublin scenario. However, the results are different from those of your paper.

image

image

As you can see, CoTVAll has the lowest performance and does not converge.

Guojyjy commented 9 months ago

The implementation of CoTVAll corresponds to CoTV* in our paper, in which all CAVs are trained as DRL agents. The significant increase in the number of DRL agents makes training convergence difficult.

CoTV can obtain good traffic performance under small-scale and synthetic road networks. Your results do not conflict with that of our paper. The 1km^2 Dublin scenario is too complex and too large-scale for CoTV.

Besides, your assigned computing resource (i.e., 2 CPUs) may not be enough for the training.

TrinhTuanHung2021 commented 9 months ago

Thank you very much for your reply. It helps me a lot to understand clearly your models.

bahramimostafa1997 commented 8 months ago

Hello Thank you for uploading your models. I have implemented your models with CoTV under the Dublin scenario. However, the results are different from those of your paper.

image

image

As you can see, CoTVAll has the lowest performance and does not converge.

hi . you run this project on windows or linux? do you know that can we install flow on windows or not? tanks

TrinhTuanHung2021 commented 8 months ago

Hello Thank you for uploading your models. I have implemented your models with CoTV under the Dublin scenario. However, the results are different from those of your paper. image image As you can see, CoTVAll has the lowest performance and does not converge.

hi . you run this project on windows or linux? do you know that can we install flow on windows or not? tanks

Hi. I run these models on Linux. We can not install flow on Windows, only on Linux