Coordination within multiple learning automata agents: a novel distributed permission switching protocol
P.S. Aggarwal, Chenhui Liu · 2006
We have implemented a reinforcement learning-based protocol for an ad hoc wireless LAN with multiple mobile stations, capable of operating under bursty traffic conditions. A learning automaton deployed at each agent (i.e. mobile station) estimates using noisy network information, as to which agent is most likely to transmit at any given instant. Using a round-robin polling method, each automaton determines when to switch its choice for the agent that may have been granted transmission permission at any given instant. The learning automaton scheme selected, the probabilistically-switch-action-on-failure automaton (PSAFA), has been designed to perform such switching optimally in nonstationary noisy environments. We show that the selected learning automaton performs better than the L/sub R,P/ scheme-based protocol. Therefore, we propose the novel learning automaton for achieving coordination among multiple agents sharing a resource with uncertain availability.