E2E Delay Reduction in VANETs using SARSA-based Reinforcement Learning

Vehicular ad hoc networks (VANETs) are integral to intelligent transportation systems, enabling real-time vehicle communication to enhance safety and traffic flow. However, the dynamic topology, frequent disconnections, and limited bandwidth in VANETs pose significant challenges for efficient routing. To address these issues, we propose a SARSA-based reinforcement learning (RL) routing strategy that adapts to network changes by learning from environmental interactions. This approach updates routing decisions based on observed state-action pairs to optimize performance. We evaluate the proposed method against established protocols such as AODV, DSR, OLSR, GPSR, and DSDV. Results show a 57% reduction in average end-to-end delay, along with notable improvements in packet delivery ratio and throughput, demonstrating the effectiveness of RL in enhancing VANET routing performance.