Folder description

generate_roadnet folder

A Python pipeline using NetworkX to convert real-world data (Manhattan, Porto and Hangzhou) from OpenStreetMap and taxi trajectories into a simulator-compatible format, including traffic flow, signals, intersections and roads data.

run folder

run_XX.py : contains training process of different baselines (partial).

actor_critic folder

Contains realizations of several deep learning and reinforcement learning algorithms (GCN, Actor-Critic network)

agent folder

Contains several agents with different behavior modes in the Multi-Agent RL environment (eg. roads, drivers).

metric folder

Contains some API calculating metrics for agents’ cost.

dataset folder

Contains roadnet files for simulated world construction.

frontend folder

Contains some fronted tools for traffic visualization.

You could use index.html and the generated .txt file. Run

python download_replay.py

to download example replay txt files after you finish the training process. Checkout Document for more instructions.