Welcome to Greenwave
Greenwave is a multi-agent contexual reinforcement learning benchmark suite build on real-world data based eco-driving application. It is designed to be a flexible and easy-to-use tool for developing and comparing contexual reinforcement learning algorithms.
Note
This project is under active development.
Why Greenwave?
Our motivation for developing Greenwave are two fold.
From reinforcement learning community point of view: While Reinforcement Learning (RL) has shown considerable progress in tackling increasingly complex tasks, many RL algorithms in multi-agent settings still struggle with even minor environmental changes, hindering their real-world applicability. Despite ongoing efforts to address this challenge, the lack of real-world grounded benchmark problems impedes fair, reliable, and reproducible comparisons of different approaches. To fill this gap, in Greenwave, we provide a million traffic scenarios stemming at signalized intersections in which the goal is to control a fleet of vehicles to achieve a fleet-level emission reduction objecvtive. Greenwave encapsulates these scenarios in their digital twins format and interfaces them as a contextual Markov Decision Process to study robustness and generalization as a contextual reinforcement learning problem.
From inteligent transportation systems community point of view: While many studies have reported algorithms for eco-driving (both multi-agent and single-agent formulations), we increasingly see the need for a benchmark suite that can be used to compare these algorithms in a fair and reproducible manner. Moreover, we observe that many studies hand-pcik the scenarios for their experiments, which may not be representative of the real-world scenarios and could lead to evaluation overfitting. Greenwave provides a million traffic scenarios stemming at signalized intersections and that are grounded in real-world data. We believe a Greenwave benchmark suite will standardize the deisgn and evaluation of eco-driving algorithms.
Check out the Usage section for further information, including how to Installation the project.