LibSignal is a library for cross-simulator comparison of reinforcement learning models in traffic signal control tasks. This library is developed to implement recent state-of-the-art reinforcement learning models with extensible interfaces and unified cross-simulator evaluation metrics.
LibSignal provides transformations across different simulators, including SUMO, CityFlow and CBEngine, enabling comparisons between different algorithms originally conducted in different simulators.
LibSignal provides unified interfaces to process and retrieve common information from different simulators for state, reward and evaluation metrics.
LibSignal implements three baseline controllers and seven RL-based controllers to form a comprehensive model warehouse. Meanwhile, LibSignal collects 9 commonly used datasets.
LibSignal enables a modular design of different components, allowing users to flexibly insert customized components into the library.
LigSignal maintains RL-based traffic signal control papers published in the recent years on top conferences and journals and their workshop papers. We will continue to update the collection.
LibSignal has an OpenAI Gym interface which allows easy deployment of standard RL algorithms and supports existing RL frameworks and different simulators.
Run in Google Colab,
or download the environment and compile the source code.