Decentralized Consensus in Robotic Swarm for Collective Collision and Avoidance

Abstract

In this paper, a decentralized consensus algorithm for a robotic swarm is presented, which enable agents to escape collisions and avoid obstacles collectively. To achieve consensus, we have used neighboring clusters and extreme function for data dissemination and consensus in agents to avoid obstacles respectively. The main challenge is to make fast and accurate collective decision by improving data propagation in presence of Byzantine agents. To improve data dissemination in local interactions of agents in decentralized fashion, an iterative rumor-based data propagation model is proposed. Due to presence of Byzantine robots, the LCP and the W-MSR algorithm cannot achieve consensus for obstacle avoidance in artificial swarms. We establish the Expectation-based Extreme Value (EEV) algorithm using the local expectation and the extreme function to solve these problems. The experiments conducted in simulations demonstrate that the rumor spreading method has better results than the Peer-to-Peer method in randomly connected swarm signaling network (SSN) with complex environmental circumstance, the EEV algorithm is more effective than the LCP and the W-MSR for the swarm navigation and consensus in agent on large obstacles / environmental features. Furthermore, in presence of malicious / hacked agents in a swarm it is very difficult to reach consensus. The result show that proposed algorithm can handle the Byzantine agents effectively.