A Multi-agent OpenAI Gym Environment for Telecom Providers Cooperation

The ever-increasing use of the Internet (streaming, Internet of things, etc.) constantly demands more connectivity, which incentivises telecommunications providers to collaborate by sharing resources to collectively increase the quality of service without deploying more infrastructure. However, to the best of our knowledge, there is no tool for testing and evaluating participation strategies in such collaborations. This article presents a new adaptable framework, based on the OpenAI Gym toolkit, allowing to generate customisable environments for cooperating on radio resources. This framework facilitates the development and comparison of agents (such as reinforcement learning agents) in a generic way. The main goal of the paper is to detail the available functionalities of our framework. We then focus on game theory aspects as multi-player games induced by these environments can be considered as sequential social dilemmas. We show in particular that although each agent has no incentive to remain cooperative at each step of such iterated games, a mutual cooperation provides better outcomes (in other words, Nash Equilibrium is non optimal)