search

multi-commander / Multi-Commander

Multi & Single Agent Reinforcement Learning for Traffic Signal Control Problem
Apache License 2.0
116 stars 30 forks source link

Traffic light baseline comparison #3

Closed ThisIsIsaac closed 5 years ago

ThisIsIsaac commented 5 years ago

Do you have any rule-based non-RL simultions that you used to gauge the performance of RL agents against?

xiawenwen49 commented 5 years ago

Yes, we have used presslight & max pressure methods, and actually they perform well. The DQN outperforms other RLs for the single intersection scenario.

Skylark0924 commented 5 years ago

You can see the code in /SingleAgent/Perfect. It is actually based on rules.

ThisIsIsaac commented 5 years ago

oh, I assume presslight is when the person in a crosswalk presses a button, and max pressure uses rule-based methods and sees the number of cars on each lane to make judgement.

There isn't a method where they take nothing as input, and only cycles through the phases?

Skylark0924 commented 5 years ago

I do not understand your question since the count of cars on each phase is already an input of MaxPressure