基於可滿足性模組理論之實數平面多智能體協同安全定位

Abstract

多智能體系統 (multi-agent system) 基於各智能體之間的溝通並合力解決問題，如智慧交通系統中的各項運輸工具，軍事系統中的各種軍事設備，機器人系統中的自主機器人等，這些智能體們在各系統中藉由合作達到單一智能體無法完成的任務。其中，定位在多智能體系統中是舉足輕重的工作之一，例如自駕車需要精準定位來達到自動駕駛。本論文考慮互聯多智能體系統的協同安全定位，並解決兩項主要問題：每個智能體有各自的定位誤差，且系統中存在刻意散佈錯誤資訊的攻擊者。針對以上問題，本論文基於可滿足性模組理論設計了建構性方法及破壞性方法，兩種方法皆嘗試使各智能體達成定位上的共識，並同時確認攻擊者存在之有無。實驗結果顯示，相較於一基準方法，建構性方法能提升攻擊者辨識的準確度，而破壞性方法可以加速取得定位共識所需時間。

Multi-agent systems are an emerging technology with a promising future, and positioning is a fundamental task supporting applications of multi-agent systems, with vehicles needing precise positioning to perform autonomous driving as a prime example. In this thesis, we consider cooperative positioning for connected multi-agent systems. In particular, we address the two following challenges: Each agent has its own positioning errors, and malicious agents within the group intentionally provide false information. Based on Satisfiability Modulo Theories (SMT), we design two approaches, a constructive approach and a destructive approach, to reach a positioning consensus among the agents and identify the set of potential attackers. Experimental results demonstrate that the proposed approaches improve the consensus accuracy and speed up the consensus process, respectively, compared with a baseline approach.