Resilient Consensus Algorithms for Distributed Control Under Adversarial Conditions: Simulation-Based Comparative Analysis

Abstract

Distributed consensus algorithms form the foundation of cooperative control in multi-agent unmanned systems, yet their vulnerability to adversarial attacks—including Byzantine agents, data injection, and communication jamming—poses critical risks for safety-critical applications. This paper presents a comprehensive simulation-based comparative analysis of resilient consensus algorithms designed to maintain coordination despite malicious or faulty agents. We examine fundamental resilience mechanisms including weighted mean subsequence reduced (W-MSR), approximate Byzantine consensus, and resilient asymptotic consensus algorithms, analyzing their theoretical guarantees regarding the maximum tolerable fraction of adversarial agents. The study develops a unified simulation framework implementing twelve prominent resilient consensus variants across representative multi-agent scenarios: distributed state estimation, cooperative target tracking, and formation control under attack. Our simulation environment models diverse adversarial strategies including value falsification, gradient attacks, and strategic message dropping, with adversary capabilities ranging from omniscient to locally-informed. Performance evaluation encompasses convergence accuracy, convergence time, communication overhead, and computational complexity as functions of network size (10-100 agents), adversary fraction (0-40%), and network topology (complete, random geometric, scale-free). Comparative results reveal that graph-based filtering approaches achieve superior resilience in dense networks, while reputation-based methods excel in sparse topologies with persistent adversaries. We investigate the fundamental trade-off between resilience guarantees and convergence speed, demonstrating that algorithms with stronger Byzantine tolerance exhibit significantly slower convergence in benign conditions. The paper examines adaptive algorithms that adjust resilience mechanisms based on detected anomalies, showing improved performance across varying threat levels. Advanced techniques including machine learning-based anomaly detection, cryptographic authentication, and blockchain-based consensus are evaluated for integration with classical resilient algorithms. Simulation analysis quantifies the impact of network topology on resilience, identifying critical connectivity thresholds below which no algorithm can guarantee consensus under adversarial conditions. The study concludes with architectural recommendations for implementing resilient consensus in real-world unmanned systems and identifies open challenges in dynamic adversary adaptation and resource-constrained environments.