The Artificial Life and Adaptive Robotics Laboratory ALAR Technical Report Series Risk Assessment of Capability Requirements Using WISDOM II Ang Yang, Hussein A. Abbass, Ruhul Sarker TR-ALAR- The Artificial Life and Adaptive Robotics Laboratory School of Information Technology and Electrical Engineering University of New South Wales Northcott Drive, Campbell, Canberra, ACT Australia Tel: + Fax:+
Risk Assessment of Capability Requirements Using WISDOM II Ang Yang, Hussein A. Abbass, Ruhul Sarker School of Info. Tech. and Electr. Eng. Univ. College, Univ. of New South Wales ADFA, Canberra ACT Australia Email: {Ang.Yang, h.abbass, r.sarker, m.barlow}@adfa.edu.au Abstract The analysis of capability requirements plays a very important role in military operations. It assists analysts to make decisions at strategic, operational and tactical levels. The analysis of capability requirements tends to be extremely expensive and time-consuming. Cutting-edge information technologies, such as red teaming, complex adaptive systems and agent based systems, can facilitate such analysis through computer simulations. Based on these technologies, a promising agent-based combat simulation system WISDOM II is built. In this paper, we conduct a series of analysis to evaluate the effect of different capability configurations on the performance of different force compositions. I. INTRODUCTION Red teaming, as a risk assessment and analysis tool, has widely been adopted in military exercises by defense organizations [], [], []. In red teaming, the RED team acts as the opponent of the BLUE team. The BLUE team attempts to find the risk in the eye of its adversary or competitor, the RED team, and the way to mitigate them. Normally there are two types of red teaming: human-based red teaming and software-based red teaming. Human-based red teaming is extremely expensive and does not enable analysts to explore all aspects of the problem [], []. Computer simulations of multi-agent systems are used for software-based red teaming. These simulations explore abstract higher level scenarios of different vulnerabilities in a plan or operation. Once the weaknesses in the system are identified and a risk analysis is conducted, human-based red teaming can be used in a more focused way to increase the fidelity of the analysis. Through red teaming, analysts can learn about the future by forward prediction of scenarios. It is easy for them to analyze capability requirements and answer questions such as: what kind of weapons are required, what kind of communication should be used, how large the team size should be, what is the right team composition, etc. In this paper, we adopt software-based red teaming and use WISDOM II [], the version II of the warfare intelligent system for dynamic optimization of missions (WISDOM)[], [], as the simulation platform. Several scenarios are developed, on which capability requirements are analyzed. This paper is organized as follows. We first describe the capability requirements in defence domain followed by a review of WISDOM II. Then we set up several experiments and analyze the result. Finally the conclusions are drawn. II. CAPABILITY REQUIREMENTS Generally speaking, capability means the ability to fulfill certain tasks. In the defense domain, as defined in DOD (Department of Defence, USA), capability [] is the ability to achieve a specified wartime objective, such as win a war, occupy a specified area, beat the intruders, destroy a target, etc. It covers four aspects: ) force structure the number of the units which form the force, the size of each unit, and the mixture of the units; ) modernization technical complexity of forces, units, weapon systems and equipments; that is, what kind of technologies, weapon systems and equipments are used by forces and units. ) unit readiness the capability displayed by units to fulfill the missions assigned to them by the combatant commander.
) sustainability the ability to maintain a certain level of military operations for a certain time in order to execute their assigned missions. It takes into accounts that all necessary supply including both substantial and spiritual should be kept at a necessary level and lasted for a certain period of time in order to execute the required missions. The substantial supply includes the size of forces, any necessary weapons available, all logistics, etc. The spiritual supply includes the moral level of individual combatants, units and the whole force. Analyzing capability requirements, planning and development are crucial issues for defense at all decision making levels: strategic, operational and tactical. The required capabilities vary at different levels [] because the objectives are different. At the national level, the objective is to maintain the social and economical order and sustaining its economy. This requires the defence force to be able to defend any attack including terrorism. However, the economic and technological power of the nation limits its capabilities. At an operational level, capabilities mean the ability for the units or battalions to effectively and efficiently fulfill military objectives assigned by their commanders. The capability needs are totally mission dependant and different missions require different capabilities. In the study of Army AS a System [], [], the work attempted to capture the essence of a force by breaking down force capabilities into seven generic core skills. These seven skills are: engagement, information collection, communication, decision making, sustainment, movement and protection. This paper focuses on three aspects: force size, engagement and communication. Scenarios with different combinations of these three aspects are developed, simulated and analyzed using WISDOM II. III. WISDOM II WISDOM II was inspired by existing agent based distillation (ABD) combat systems. Our recent research [], [] shows that most existing ABDs [], [], [], [], including WISDOM version I [], [], were developed mainly on platform centric (traditional) warfare and current agent architectures, which limit their ability to study network centric warfare (NCW). WISDOM II was re-designed and re-developed based on a new agent architecture, called Network centric multi-agent architecture (NCMAA) []. NCMAA is purely based on network theory [], [], [], []. The system is designed on the concept of networks, where each operational entity in the system is either a network or a part of a network. The engine of the simulation is also designed around the concept of networks. Generally speaking, there are five distinct components in WISDOM II. The first three components are used to model the internal behavior of warfare, while the last two are used for analysis. ) the C component including both command and control (C), and communication. ) the sensor component retrieving information from the environment ) the engagement component including firing and movement activities ) the visualization component presenting various information with graphs ) the reasoning component interpreting the results in natural language during the simulation process Five types of networks are defined in WISDOM II; these are the command & Control (C), vision, communication, information fusion, and engagement networks. Four types of agents are supported in WISDOM II: combatant agent, group leader, team leader and swarm leader. Agents are defined by their characteristics and personalities. Each agent has nine types of characteristics: health, skill, probability to follow command, visibility, vision, communication, movement and engagement. Initial levels of health, skill, visibility and vision are predefined by the user. They may be different for different agent types. The swarm leader can also form plans and give orders to combat groups. The personality in WISDOM-II is defined by two values: a magnitude and a direction vector representing the attraction-repulsion direction and weight for each agent. The movement of each agent is determined by its situation awareness and personality vector. In each time step, the agent can only move to its neighbor cells based on the overall influence of all perceived agents. A strategic decision is made by the swarm leader of each force based on the common operating picture (COP), which is the global view of the battle field for that force. Decision making on the force level utilizes the same environment the agents are embedded in, but on a coarser resolution. WISDOM II collects information for each entity as well as for the interaction between entities. In this way, a large number of statistics are collected. Then these statistics are fed into a reasoning engine, where natural language
reasoning is provided to the user. WISDOM II also provides capabilities such as interactive simulation. For more details of WISDOM-II, please refer to []. WISDOM II overcomes the following drawbacks of existing ABDs. Hard to validate and verify New system behaviors emerging from simple low level rules is one essential characteristic for any complex adaptive system (CAS). In existing ABDs, agents are programmed without an underlying theoretically sound software architecture. Therefore it is very difficult to validate and verify them. However in WISDOM II, all agents are modelled on the concept of network. By examining the topology or other characteristics of network, the developers or users are easily to identify the errors if something goes wrong within the system. No reasoning during the simulation The critical drawback of existing ABDs is the lack of reasoning during the simulation. Due to the high degree of nonlinear interaction between agents, it is almost impossible to reason at the agent level, which makes it harder to understand the results of the whole simulation. The architecture of NCMAA make WISDOM-II more suitable for conducting reasoning on the network (group) level. The reasoning component is defined through two stages () establishing causal relationships; and () establishing degree of influence. The first stage is taken care of through the influence diagram, which defines the causal relationship. The second stage is undertaken through statistical inferencing. The causal network identifies the influence between agents interactions, two kinds of reasoning are conducted in each time step: time series analysis and correlation analysis between network measures and military measures. Correlation analysis may allow analysts to understand which network is playing the key role in the outcome and to see why certain behaviors emerge during the simulation while the time series analysis helps analysts to understand the dynamics over time. Finally the result is presented to the user in natural language. No connection between tactic and strategic levels Existing ABDs are developed either on the reactive agent architecture [], [], [], [] which focuses on tactics, or on the BDI (belief-desire-intention) [], [], [], [], [] architecture which focuses on strategies. There is almost no interaction between tactics and strategies being modelled by existing ABDs. WISDOM II employs two levels of decision making mechanisms: tactical and strategic, which allow analysts to study the influence between these two levels. Hard to capture the underlying structural interaction between agents Although existing other ABDs embed the structural interaction between agents, there is no explicit model for such interactions. It is hard for the user to capture these interactions during the simulation, which is a crucial point in a CAS. WISDOM II is developed on the architecture of NCMAA, which is purely based on network theory. Each interaction between the agents is presented as a network. Studying the role of each network in the simulation will ease the understanding of important interactions. Difficulty in application to complexity Current ABDs are based on conventional military tactics and tend not to be approached from an overarching system view. Concepts such as NCW, with its inherent complexity and interdependency, present challenges to identifying correct inputs at the entity level. Thus, techniques addressing higher level manipulations must be employed. IV. SCENARIO SETUP In order to identify the role of force size, engagement and communication in military operations in different environments, three typical environments are defined in Figure. Scenario a (on left) represents an open environment without any obstacles, Scenario b (middle figure) represents a natural environment with random obstacles while scenario c (right figure) represents an environment with organized obstacles. Each environment is a x grid.
Fig.. The three different environments used in this paper. For each environment, the red force with fixed capabilities plays against the blue force with different capabilities. The characteristics of both forces are shown in Table I. In this paper, a platform centric force means that each agent may communicate with any other agents if they are within each other communication range while network centric force means that each agent can only send information to the force headquarter and then the common operation picture (COP) developed by the force headquarter will be sent back to each agent in the battlefield. The lost probability is the probability that messages are lost in a communication channel. For both blue and red agents, they will lose mobility after hits and die after hits. TABLE I FORCE CHARACTERISTICS Blue Force Red Force Force structure Platform or network centric Platform centric Force size Agents Agents Sensor range Communication range Lost probability % Weapon system direct weapon direct weapon Firing range Personality Attempt to attack the red agents Run away from the blue agent V. RESULT AND ANALYSIS For each setup, we run the simulation times and use the average loss exchange ratio to measure the performance of the blue force. In order to avoid division by zero, the loss exchange ratio is calculated as in Equation. R = Red Casualty + Blue Casualty Figures,, and depict the difference between the average loss exchange ratio between an NCW force and a platform force. A number of observations can be made from these figures. By inspecting environment a, the more blue agents we have, the less the advantage of NCW over platform forces. One can also see that the advantages of NCW diminishes as the probability of message losses in the communication channels increases. The advantages of NCW also diminishes when the environment is full of obstacles. Naturally, maneuvers in environment b is more difficult than environment a. In addition, it seems that when the obstacles are distributed as a maze, and as the number of blue agents increase, the platform centric setup seems to outperform the NCW setup. As such, when forces are deployed in an urban environment with lots of mazes (such as fighting in the streets of a busy city), NCW is not an advantage. Figures,, and show the interaction between the communication range and the fire range. It is clear that in an open terrain, a long communication and fire ranges emerged as an advantage with the loss exchange ratio almost ()
..................................................................................... Fig.. Environment a: The difference between the loss exchange ration in an NCW and a platform setting for the blue force. The graph top-down corresponds to force size of to agents in a step of, respectively.
Fig.. Environment a: The loss exchange ration in a platform setting for the blue force contrasting the effect of communication and fire ranges. The graph top-down corresponds to force size of to agents in a step of, respectively.
..................................................................................... Fig.. Environment b: The difference between the loss exchange ration in an NCW and a platform setting for the blue force. The graph top-down corresponds to force size of to agents in a step of, respectively.
Fig.. Environment b: The loss exchange ration in a platform setting for the blue force contrasting the effect of communication and fire ranges. The graph top-down corresponds to force size of to agents in a step of, respectively.
................................................................................ Fig.. Environment c: The difference between the loss exchange ration in an NCW and a platform setting for the blue force. The graph top-down corresponds to force size of to agents in a step of, respectively.
Fig.. Environment c: The loss exchange ration in a platform setting for the blue force contrasting the effect of communication and fire ranges. The graph top-down corresponds to force size of to agents in a step of, respectively.
approaches in the case of blue agents. However, as the complexity of the environment increases, the loss exchange ratio does not exceed as in the case of environment c with agents. It is also clear from all figures that the blue agent should more capabilities than the red team so that the loss exchange ratio becomes greater than one. VI. CONCLUSION AND FUTURE WORK Capability requirements, planning and development are very important for defense. A good tool may facilitate this kind of analysis. This paper adopts WISDOM-II as a simulation platform and evaluates the role of force size, communication and engagement in the force performance. Three setups are developed to simulate three typical real scenarios: open plain environment, natural field environment and complex urban environment. As expected, the performance improves when the force size increases, the communication quality improves or the firing range increases. However, the obstacles may hinder such an improvement if the force is only equipped with direct weapons. In order to obtain more combat power from information superiority, the force should have matching weapon system, e.g. long range firing weapon or indirect weapon. Our analysis also suggests that the efficient local communication is less important within an environment of lots of obstacles. This paper opens a number of questions for our future research. For example, identifying the effect of indirect weapon systems; and the effect of the level of clustering of the agents in the environment. In the future, we would also like to conduct analysis on other capabilities, e.g. information collection, movement, protection, etc. ACKNOWLEDGMENTS This work is supported by the University of New South Wales grant PS and the Australian Research Council (ARC) Centre for Complex Systems grant number CEO. The authors also wish to thank Dr. Michael Barlow from School of ITEE, UNSW@ADFA for useful discussions and Dr.Yin Shan from School of ITEE, UNSW@ADFA for constructive comments. REFERENCES [] DOD, Defense Science Board Task Force on The Role and Status of DoD Red Teaming Activities, Tech. Rep. Unclassified Report -, Office of the Under Secretary of Defense For Acquisition, Technology, and Logistics Washington, D.C.,. [] J. F. Sandoz, Red teaming: a means to military transformation, IDA Paper P-, Institute for Defense Analyses, January. [] A. Yang and H. A. Abbass and R. Sarker, Characterizing Warfare in Red Teaming, IEEE Transactions on Systems, Man, Cybernetics, Part B: Cybernetics (),. [] A. Yang and H. A. Abbass and R. Sarker, Landscape Dynamics in Multi-agent Simulation Combat Systems, in Proceedings of th Joint Australian Conference on Artificial Intelligence, LNCS, Springer-Verlag, (Cairns, Australia), December. [] A. Yang, Understanding Network Centric Warfare, ASOR BULLETIN, pp., December. [] A. Yang and H. A. Abbass and R. Sarker, WISDOM-II: A Network Centric Model for Warfare, in Ninth International Conference on Knowledge-Based Intelligent Information & Engineering Systems (KES ), LNCS, (Melbourne, Australia), September. [] DOD, Department of Defense Dictionary of Military and Associated Terms. http://www.dtic.mil/doctrine/jel/doddict/,. Access on Sept.,. [] R. Gori and P. Chen and A. Pozgay, Model-Based Military Scenario Management for Defence Capability, in the proceedings of the th International Command and Control Research and Technology Symposium, (Copenhagen, Denmark), September -. [] N. J. Curtis and P. J. Dortmans, A dynamic conceptual model to explore technology-based perturbations to a complex system: the land force, Asia-Pacific Journal of Operational Research (), pp.,. [] D. R. Shine, An exploratory study of the arm-as-a-system core skills, Journal of Battlefield Technology (), pp.,. [] A. Ilachinski, Irreducible Semi-Autonomous Adaptive Combat (ISAAC): An Artificial Life Approach to Land Combat, Research Memorandum CRM -, Center for Naval Analyses, Alexandria,. [] A. Ilachinski, Irreducible Semi-Autonomous Adaptive combat (ISAAC): An Artificial Life Approach to Land Combat, Military Operations Research (), pp.,. [] M. K. Lauren, Modelling Combat Using Fractals and the Statistics of Scaling Systems, Military Operations research (), pp.,. [] M. Barlow and A. Easton, CROCADILE - An Open, Extensible Agent-Based Distillation Engine, Information & Security (), pp.,. [] S. Wasserman and K. Faust, Social Network Analysis: Methods and Applications., Cambridge University Press, Cambridge, UK,. [] Rka Albert and Albert-Lszl Barabsi, Statistical mechanics of complex networks, Reviews of Modern Physics (), pp.,. [] M. E. J. Newman, The structure and function of complex networks, SIAM Review (), pp.,. [] S. N. Dorogovtsev and J. F. F. Mendes, Evolution of networks, Advances in Physics (), pp.,.
[] M. J. Wooldridge and N. R. Jennings, Intelligent Agents: Theory and Practice, Knowledge Engineering Review, pp., June. [] H. S. Nwana, Software Agents: An Overview, Knowledge Engineering Review (), pp.,. [] K. P. Sycara, Multiagent Systems, AI Magazine (), pp.,. [] M. J. Wooldridge, Intelligent Agents, in Multiagent Systems: A Modern Approach to Distributed Artificial Intelligence, G. Weiss, ed., ch., pp., The MIT Press,. [] A. S. Rao and M. P. Georgeff, BDI Agents: From Theory to Practice, in Proceedings of the st International Conference on Multi-Agent Systems, pp., (San Francisco, USA),.