Distributed learning automata based data dissemination in swarm robotic systems

Abstract

Swarm robotics systems often work in collaboration with humans to accomplish tasks in a random environment. Swarms allow many tasks to be accomplished within a mission more quickly while generally being more cost effective. The robots in the swarm can also accomplish tasks that humans are not able to perform on their own. The random environments the swarms work in render any previous contact data between robots useless as the contact patterns are different for each deployment. In the case of military and disaster scenarios, delivering data items quickly is imperative to the success of a mission. However, robots have limited battery and need a lightweight protocol that maximizes data delivery ratio and minimizes data delivery latency while consuming minimal energy. Learning automata are a form of Reinforcement Learning that is computationally inexpensive and can adapt to a dynamic environment. This combination allows for lightweight decision making based on the current topology of the network. We present two learning automata based data dissemination protocols, LADD and sc-LADD. LADD uses learning automata with direct connections to all neighboring nodes to make efficient and accurate forwarding decisions while sc-LADD uses learning automata and exploits the swarming nature of the robotic systems to abstract swarms and reduce the number of decisions available to the learning automata, which also reduces overhead. Results from our extensive simulation in NS3 indicate that these two protocols complement each other to achieve these goals: LADD can be used to maximize data delivery ratio when residual energy is high, while sc-LADD can be used to significantly reduce overhead and maintains a reasonable delivery ratio when residual energy is low. Furthermore, simulation results are verified as they conform to our theoretical analysis of packet delay and learning automata convergence.