Multi-Agent Reinforcement Learning Based Efficient Routing in Opportunistic Networks
Jagdeep Singh, Sanjay Kumar Dhurandher, Isaac Woungang · 2020
In this paper, a novel Multi-Agent Reinforcement Learning Congestion Control (MARL-CC) routing protocol for opportunistic networks is proposed, in which no prior knowledge about the target environment (i.e. the network) is assumed and the only way to acquire the information about this environment is by interacting with it through continuous online learning. In the proposed scheme, a node gets its input (such as buffer occupancy, set of neighbors nodes, to name a few) from the environment. Based on this, it chooses an action to take from a set of possible actions. Depending on this action's effectiveness in controlling the congestion, the node that has issued it is given a reward or a penalty; the overall goal being to maximize the reward (or minimize the congestion). Simulation results show that our proposed scheme outperforms the RLPRoPHET and PRoPHET routing schemes in terms of delivery probability and overhead ratio.