Reinforcement learning as adaptive network routing of mobile agents

In large, distributed systems, like ad-hoc networks, centralized learning of routing or movement policies may be impractical. We need to employ learning algorithms that can learn independently, without the need for extensive coordination. A search for alternative methods of routing packets has resulted in reinforcement learning (RL) as a good approach to adaptive routing. RL methods are able to learn and adapt to a unknown and changing environment. Using only a simple coordination signals such as a global reward value, we show that RL methods can be used to control routing of mobile agents. RL method was used to regulate the transfer of mobile agent from the network input, through the routing nodes, towards the service processing nodes. A distributed Q-Learning framework, based on RL, embeds a learning policy at every node to adapt itself to the changing network conditions, which leads to a synchronized routing information, in order to achieve a shortest delivery and service processing time of mobile agents.