Gossip based asynchronous and randomized distributed task assignment with guaranteed performance on heterogeneous networks

Abstract

The main contribution of this paper is a novel distributed algorithm based on asynchronous and randomized local interactions, i.e., gossip based, for task assignment on heterogeneous networks. We consider a set of tasks with heterogeneous cost to be assigned to a set of nodes with heterogeneous execution speed and interconnected by a network with unknown topology represented by an undirected graph. Our objective is to minimize the execution time of the set of tasks by the networked system. We propose a local interaction rule which allows the nodes of a network to cooperatively assign tasks among themselves with a guaranteed performance with respect to the optimal assignment exploiting a gossip based randomized interaction scheme. We first characterize the convergence properties of the proposed approach, then we propose an edge selection process and a distributed embedded stop criterion to terminate communications, not only task exchanges, while keeping the performance guarantee. Numerical simulations are finally presented to corroborate the theoretical results.

Replication-Based Data Dissemination in Connected Internet of Vehicles

“…To study network consensus in strongly connected networks, Wu et al [30] presented a gossip-based algorithm and showed it could quickly reach consensus as well as reducing the consumed transmissions. Franceschelli et al [31] studied the execution time of heterogeneous tasks in an undirected graph. They proposed randomised interaction algorithm based on gossip to let the nodes cooperatively complete the tasks to minimise the task execution time.…”

Robot Planning Based on Boolean Specifications Using Petri Net Models

“…Other related discrete path planning solutions are presented in [7], [8], where the authors assume tasks combining Boolean variables on the graph nodes and define a new language as an extension of the Generalized Traveling Salesman Problem for final system states, and in [15], [16], [17], where MILP (Mixed ILP) techniques were proposed for solving different planning or allocation problems in a centralized or distributed fashion. With respect to these works, our method allows Boolean specifications also on robot trajectories, not only on the system final states.…”