A Survey of Multi-robot Regular and Adversarial Patrolling Li Huang, Student Member, IEEE, MengChu Zhou, Fellow, IEEE, Kuangrong Hao, Member, IEEE, and Edwin Hou, Senior Member, IEEE Abstract — Multi-robot systems can be applied to patrol a con- cerned environment for security purposes. According to different goals, this work reviews the existing researches in a multi-robot patrolling field from the perspectives of regular and adversarial patrolling. Regular patrolling requires robots to visit important locations as frequently as possible and a series of deterministic strategies are proposed, while adversarial one focuses on unpre- dictable robots’ moving patterns to maximize adversary detec- tion probability. Under each category, a systematic survey is done including problem statements and modeling, patrolling objectives and evaluation criteria, and representative patrolling strategies and approaches. Existing problems and open questions are presented accordingly. Index Terms—Multi-robot systems, regular patrolling, adversarial patrolling, coordination methods, surveillance. I. Introduction M ULTI-ROBOT systems (MRSs) are composed of mul- tiple robots (including mobile robots) where they com- municate and coordinate with each other explicitly or impli- citly to execute tasks efficiently. Thus, these robots form a team. Considering the general characteristics and potential ad- vantages of MRSs [1], many regard MRSs as one of the most promising strategies in the field of policing crimes [2], [3], monitoring incidents [4] and managing disasters [4]–[7]. Compared with wireless sensor networks constructed for en- vironmental surveillance and information perception, MRSs are less costly on facility construction and maintenance, are more flexible to be applied in emergent situations and can ex- ecute tasks with equipped devices if needed. Patrolling is a fundamental task in the above-mentioned applications, which is known as the activity of moving around in an area to protect or supervise it for security purposes [8]. Since it was first comprehensively addressed in the MRSs field in 2002 [9], a number of research achievements overcoming different difficulties and focusing on various key points are made. A review of them may contribute to the deep understanding of the past work and support for future studies. Existing survey papers in a multi-robot patrolling field have been done by Almeida et al. [10] in 2004 and Portugal et al. [11] in 2011. In [10], Almeida et al. give an overview of the approaches investigated by their research groups. The strategies are summarized based on three techniques: Traveling Salesman Problem-based single-cycle approach, negotiation mechanism and reinforcement learning. In [11], Portugal et al. introduce 17 proposed strategies along with the development of multi-robot patrolling studies from 2002 to 2010 and summarize their main characteristics from the perspectives of agent type, perception, communication, coordination and decision-making mechanisms. Both papers show that a good strategy should minimize the time lag between visits at specific locations of an environment. This principle leads to deterministic patrolling routes and periodic visiting frequency. When it comes to a scenario with intruders attempting to penetrate an area or steal jewelry from a safe box, they can easily obtain a full knowledge of the patrolling strategies after a period of observation and achieve their nefarious goals at an interval between two consecutive visits of the robots. For this case, the maximum visiting frequency becomes less crucial and the maximum probability of detecting intruders is regarded to be a main criterion, i.e., non- deterministic patrol behaviors should be studied. Therefore, we summarize the existing literature according to different patrolling goals, regular patrolling and adversarial patrolling, and make a systematic review under each category. The differences between regular patrolling and adversarial patrolling are outlined as follows: 1. Patrolling scenario: finding something unusual vs. defending against intruders Regular patrolling can be used as a daily surveillance of an environment. All places in the area need to be visited continuously such that information in the environment can be collected or whether there are accidents or something unusual can be assured. Adversarial patrolling is used as a security surveillance to prevent an environment from being invaded, polluted and damaged by intruders/offenders. It should consider their intentions and strategies. 2. Patrolling goal: minimizing the time lag between two visits vs. maximizing the detection probability of intruders/ This work was supported in part by the International Collaborative Project of the Shanghai Committee of Science and Technology (16510711100), Na- tional Natural Science Foundation of China (61603090, 61806051), the Fun- damental Research Funds for the Central Universities (2232017D-08, 2232017D-13), Shanghai Sailing Program (17YF1426100) and by FDCT (Fundo para o Desenvolvimento das Ciencias e da Tecnologia) (119/2014/A3). Recommended by Associate Editor Choon Ki Ahn. (Corres- ponding author: MengChu Zhou.) Citation: L. Huang, M. C. Zhou, K. R. Hao, and E. Hou, “A survey of multi-robot regular and adversarial patrolling,” IEEE/CAA J. Autom. Sinica, vol. 6, no. 4, pp. 894–903, Jul. 2019. L. Huang and K. R. Hao are with the Engineering Research Center of Di- gitized Textile and Apparel Technology, Ministry of Education, and the Col- lege of Information Science and Technology, Donghua University, Shanghai 201620, China (e-mail: huanglili622@126.com; krhao@dhu.edu.cn). M. C. Zhou and E. Hou are with the Department of Electrical and Com- puter Engineering, New Jersey Institute of Technology, Newark, NJ 07032 USA (e-mail: zhou@njit.edu; hou@njit.edu). M. C. Zhou is also with the In- stitute of Systems Engineering, Macau University of Science and Technology, Macau 999078, China. Digital Object Identifier 10.1109/JAS.2019.1911537 894 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 6, NO. 4, JULY 2019