GrGym: A Playground for Research on RL/AI Enhanced Wireless Networks

Abstract:
The provision of a wide range of services each with different requirements makes next generation wireless networks become more complex and heterogeneous which is aimed to be tackled through network softwarization and the application of Artificial Intelligence (AI)-based methods. Specifically, AI methods based on Deep Reinforcement Learning (RL) became very popular as they enable closed-loop end-to-end network optimization even of complex and heterogeneous wireless networks. However, for early deployments there is a pressing need for well-defined environments so that deep RL-based solutions can be studied. We present GrGym, a software framework for the development of deep RL enhanced wireless networks, with a specific focus on its usage in experimental 5G/6G research. It is based on the OpenAI Gym toolkit and the flexible GNU Radio platform. With GrGym, deep RL-based solutions for 5G/6G networks can be trained in simulated environments as well as real-world testbeds using software-defined radios.