A reinforcement learning approach to network routing based on adaptive learning rates and route memory

Mobile ad hoc networks may dynamically change their topology and system parameters. They require efficient routing techniques that provide reasonably low delivery times for packets even under high loads. A routing algorithm based on Full Echo Q-routing scheme is proposed. It uses adaptive learning rates and route memory to reduce instability of routing under high load conditions and improve performance in terms of overshoot and settling time of the learning.