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_XX.py : contains training process of different baselines (partial).
Contains realizations of several deep learning and reinforcement learning algorithms (GCN, Actor-Critic network)
Contains several agents with different behavior modes in the Multi-Agent RL environment (eg. roads, drivers).
Contains some API calculating metrics for agents’ cost.
Contains roadnet files for simulated world construction.
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.