Direct Reinforcement Learning for Autonomous Power Configuration and Control in Wireless Networks

In this paper, non deterministic Direct Reinforcement Learning (RL) for controlling the transmission times and power of a Wireless Sensor Network (WSN) node is presented. RL allows for truly autonomous optimal behaviour of agents by requiring no models or supervision to learn. Optimal actions are learnt by repeated interactions with the environment. Performance results are presented for Monte Carlo, TD0 and TDlambda. The resultant optimal learned policies are shown to out perform static power control in a stochastic environment.