DREAM-ON GYM: A Deep Reinforcement Learning Environment for Next-Gen Optical Networks

A novel open-source toolkit for a straightforward implementation of deep reinforcement learning (DRL) techniques to address any resource allocation problem in current and future optical network architectures is presented. The tool follows OpenAI GYMNASIUM guidelines, presenting a versatile framework adaptable to any optical network architecture. Our tool is compatible with the Stable Baseline library, allowing the use of any agent available in the literature or created by the software user. For the training and testing process, we adapted the Flex Net Sim Simulator to be compatible with our toolkit. Using three agents from the Stable Baselines library, we exemplify our framework performance to demonstrate the tool’s overall architecture and assess its functionality. Results demonstrate how easily and consistently our tool can solve optical network resource allocation challenges using just a few lines of code applying Deep Reinforcement Learning techniques and ad-hoc heuristics algorithms.