We are pleased to announce the release of the ns3-gym application in the ns-3 App Store, enabling ns-3 to be integrated with OpenAI Gym, and hence turn it into a playground for Reinforcement Learning in networking research. ns-3 and OpenAI Gym have become standards for academic and industry research in networking and machine learning areas, respectively. ns3-gym combines the advantages of these two; namely, the verified models of ns-3 and the simplicity of prototyping RL agents using Python libraries (e.g. Tensorflow and Keras). Its main goal is to encourage researchers from both areas to collaborate in order speed up research and development of RL-based control algorithms in networking problems.

The application is generic and can be used in various networking and communication areas. We provide the first set of problems along with baseline solutions that can serve as a tutorial to the community. A more detailed description can be found in this publication.

The ns3-gym was developed and contributed by TKN TUB team, thanks to Piotr Gawłowicz, Anatolij Zubow, and Mikołaj Chwalisz.