search

maunhb / sumo-marl

3 stars 0 forks source link

Difference between Cooperated QN and Independent QN #2

Open TrinhTuanHung2021 opened 2 years ago

TrinhTuanHung2021 commented 2 years ago

Thank you for sharing your code

But I wonder about the difference between Cooperated DQN and Independent DQN.

Could you show me clearly? or do you have any papers related your models?

maunhb commented 2 years ago

The cooperated DQN is based on Coordinated Reinforcement Learning (https://tinyurl.com/5ya6xydd). There is a paper applying it to traffic lights using DQN here: https://tinyurl.com/bdzjubs9. The Independent DQN implements a separate DQN agent at each traffic light junction and there is no coordination between them.

TrinhTuanHung2021 commented 2 years ago

Thank you for your reply to me

But i only run well static model. python experiments/static_single-intersection.py -tripfile outputs/qtrip.xml OK

And I had errors when running python experiments/coord_2x2intersection.py -tripfile outputs/qtrip.xml

traci.simulation.getCurrentTime()/1000) Traceback (most recent call last): File "experiments/coord_2x2intersection.py", line 174, in reward=r['{}'.format(edges[i])]+r['{}'.format(edges[i+1])] KeyError: '1.0' Error: tcpip::Socket::recvAndCheck @ recv: peer shutdown0 ACT 13 BUF 0)
Quitting (on error)