3neutronstar / rl-constraints-traffic

Decentralized Deep Reinforcement Learning based Real-World Applicable Traffic Signal Optimization

RL-constraints-traffic

Decentralized Deep Reinforcement Learning based Real-World Applicable Traffic Signal Optimization

Decentralized DQN

Prerequisite

• python 3.7.9 above
• pytorch 1.7.1 above
• tensorboard 2.0.0 above

How to use

check the condition state (throughput)

    python run.py simulate --network [5x5grid, 5x5grid_v2, dunsan, dunsan_v2]

"Traffic data of Dunsan and Dunsan_v2 are classified by government of South Korea."

Run in RL algorithm DQN (default device: cpu)

    python run.py train --network [5x5grid, 5x5grid_v2, dunsan, dunsan_v2]

"Traffic data of Dunsan and Dunsan_v2 are classified by government of South Korea."

• check the result Tensorboard

    tensorboard --logdir ./training_data

Hyperparameter in json, model is in ./training_data/[time you run]/model directory.

• replay the model

    python run.py test --replay_name /replay_data in training_data dir/ --replay_epoch NUM

Performance

Synthetic Data in 5x5grid(Straight Flow), 5x5grid_v2(Random Trips)

Evaluation Metric	Method	Straight Flow	Random Trips
Throughput(vehs)	Fixed	15184	6872
Throughput(vehs)	Ours	15380	7154
Waiting Time(s)	Fixed	916.7	670.2
Waiting Time(s)	Ours	851.2	587.1

Real World Data in Dunsan-dong, Daejeon, Korea

Evaluation Metric	Method	Peak	Free and Peak combined
Throughput(vehs)	Fixed	13616	10415
Throughput(vehs)	Ours	13649	10539
Waiting Time(s)	Fixed	240.1	237.6
Waiting Time(s)	Ours	231.1	218.1