ProphetRouterRL: code, simulation setup and results

Abstract

Code, simulation setup and results accompanying the paper "ProphetRouterRL: A Reinforcement Learning Enhanced Routing Protocol for V2X Delay Tolerant Networks" (submitted to MDPI Network). ProphetRouterRL keeps PRoPHET's delivery-predictability comparison as a first filter and adds a small tabular Q-learning agent that decides, for each remaining relay opportunity, whether to send the message or hold it. The agent sees six values: the message's hop count, the relay's free buffer, the remaining TTL, whether the relay is the destination, whether it has met the destination before, and its node type. The archive contains:- the Java router for the ONE simulator and the Python Q-learning agent;- ONE settings files for the plain PRoPHET baseline and for ProphetRouterRL (7-day and 12-hour runs), and the map files of the UKM-Bangi area in Malaysia, derived from OpenStreetMap;- the saved Q-tables and the agent's statistics logs;- the simulator reports and per-message result tables for the seven-day comparison and the 12-hour validation runs. The scenario has 129 nodes: pedestrians, cyclists, cars, buses, metro trains and static hubs. See README.md for the folder layout and for how to rerun the experiments, and LICENSE for the licences of the individual parts.