Learning-enforced time domain routing to mobile sinks in wireless sensor fields

We propose a learning-based approach to efficiently and reliably route data to a mobile sink in a wireless sensor field. Specifically, we consider a mobile sink that does not know when to query or does not need to query. Furthermore, the sink moves in a certain pattern within the sensor field. Such a sink passively listens for incoming data that distant source sensors unilaterally push towards it. Unlike traditional routing mechanisms, our technique takes the time-domain explicitly into account, with each node involved making the decision "at this time what is the best way to forward the packet to the sink?". In the presented scheme, motes (nodes in the vicinity of the sink) learn its movement pattern over time and statistically characterize it as a probability distribution function. Having obtained this information at the motes, our scheme uses reinforcement learning to locate the sink efficiently at any point of time.