Reinforcement Learning for Dynamic Routing in Communication Networks

Dynamic routing in communication networks has traditionally relied on heuristic or rule-based algorithms such as OSPF and BGP. However, with the rapid increase in network scale, heterogeneity, and traffic volatility, static routing policies fail to achieve optimal performance under dynamic conditions. Reinforcement Learning (RL) offers an adaptive framework for network control by allowing agents to learn routing policies through interaction with the environment. This paper presents a simulation-based analysis of RL-driven dynamic routing strategies and compares their performance with classical algorithms under varying traffic loads. The proposed study leverages Deep Reinforcement Learning (DRL) architectures in a Software-Defined Networking (SDN) environment, demonstrating improvements in average latency, packet delivery ratio, and throughput.