National Academies Press: OpenBook
« Previous: 7 Situation Awareness
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

8
Planning

Effective planning is often the key determinant of mission success in the military, and conversely, mission failure can often be traced back to poor planning. Planning's proper representation (both good and bad) within a military human behavior representation is therefore essential for the generation of realistic behaviors in the battlespace. This chapter reviews military planning in some detail, describing both doctrinal and nondoctrinal behaviors and their implications for model development. It then examines approaches taken to modeling planning—first in military planning models, and then in the work of the artificial intelligence and behavioral science communities. The chapter concludes with conclusions and goals in the area of planning.

PLANNING AND ITS ROLE IN TACTICAL DECISION MAKING

In an attempt to identify key implications for modeling the tactical planning process, this section describes the planning process itself in the context of the vignette presented in Chapter 2. Doing so allows us to identify essential attributes of this type of doctrinally correct planning behavior. This discussion is then balanced by an examination of observed planning behavior, which differs significantly from that prescribed by doctrine. The section concludes with a summary of the modeling implications of both doctrinally correct and observed planning behavior as context for the ensuing discussion of existing models and potential development approaches.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

The Tactical Planning Process

A plan can be defined as ''any detailed method, formulated beforehand, for doing or making something" (Guralnik, 1986). More specifically, an activity plan (for doing something) usually begins with a system at some initial state, specifies some desired final or goal state, and identifies constraints on the allowable sequence of actions that will take the system from the initial to final state. A valid plan is one that specifies an allowable action sequence (formulated beforehand). If it does no more than this, it is sufficing; if it optimizes some utility function of the actions and the states, it is optimizing.

Planning, or the generation of a plan, is critical to successful operations—it plays a key role in the tactical decision making process across all services and throughout all echelons. Capturing the substance of the planning process in a realistic human behavior representation is therefore essential to developing behavioral models that realistically reflect actual tactical decision making behavior.

Because of the limited scope of the present study, the focus of this chapter is limited to the U.S. Army planning process. Based on brief reviews presented to the panel by the other services, we believe the basic activities comprising this process are similar across the services, although they have different names, so the findings presented here should be generalizable. We believe similar generalization holds across echelons, although we will have more to say about this below.

The vignette in Chapter 2 describing the planning and execution of a hasty defense operation by a tank platoon incorporates a planning process that closely follows the doctrinally specified process detailed in Army publication FM 101-5, Staff Organization and Operation, and described more tutorially in Command and General Staff College publication ST 100-9, The Tactical Decision making Process. As described in the latter publication, planning is part of a five-stage process:

  1. Mission analysis

  2. Intelligence preparation of the battlefield

  3. Development of courses of action

  4. Analysis of courses of action

  5. Decision and execution

The paragraphs below provide a brief description of each of these stages. More complete descriptions can be found in FM 101-5 and ST 100-9.

The mission analysis stage begins with receipt of an order from the unit's command and proceeds to more complete definition of the initial state (or, equivalently, current situation), as well as a definition of the final goal state (or, equivalently, the mission objectives). Consideration is also given to operational constraints that will apply during the course of the operation. In the platoon-level vignette of Chapter 2, this process is formalized by expanding the mission, enemy, terrain, troops, time available (METT-T) process, which makes explicit the

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

TABLE 8.1 Relating METT-T Components to Initial and Goal States of a Plan

METT-T

Initial State

Goal State

Mission

 

X

Enemy

X

 

Terrain

X

 

Troops

X

 

Time

 

X

consideration of initial and final state planning components (see Table 8.1). This stage thus includes situation assessment activities (initial state specification) and planning activities (goal state specification).

The intelligence preparation of the battlefield stage then focuses on a detailed assessment of the situation, covering three key components (FM 101-5):

  • Environment in the area of operations (e.g., terrain, weather)

  • Enemy situation (e.g., disposition, composition, strength)

  • Friendly situation

In our platoon-level vignette, the environmental assessment is formalized by expanding the observation, cover and concealment, obstacles, key terrain, and avenues of approach (OCOKA) process, with consideration also given to the weather. Assessment of the enemy and friendly situations attempts to go beyond simple disposition and strength estimates (recall Endsley's [1995] level 2 situation awareness as described in Chapter 7) and to focus on expectations of operational behavior (level 3 situation awareness). As noted in ST 100-9, the OCOKA process may be adequate for the brigade level and below, but a more sophisticated preparation-of-the-battlefield process is typically undertaken at higher echelons based on current tactical intelligence.

The course-of-action development stage is the generative stage of the planning process,1 in which alternative sufficing plans are generated to accomplish the mission and take the unit from its current to its goal state. Each course of action is a candidate plan developed at a relatively high level, addressing such things as the type of operation and when it will happen. In our vignette, the results of this stage, which are typical, are three candidate courses of action; Army doctrine calls for the generation of "several suitable" courses of action for every enemy course of action under consideration.

The course-of-action analysis stage is the evaluative stage of the planning process,2 in which candidate courses of action are elaborated, wargamed against likely enemy courses of action (through mental simulations), and

1  

Designated the "art" of command by ST 100-9.

2  

Designated the "science" of control by ST 100-9.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
evaluated across multiple dimensions. Courses of action are then scripted out in more detail, showing (through a synchronization matrix) the anticipated temporal sequence of subordinate, support, and enemy unit activities. The wargaming itself is fairly shallow in terms of blue moves and red countermoves, covering a three-step sequence of action, reaction, and counteraction. Evaluation criteria for each course of action are specified by each commander, and each course of action is scored on the criteria according to the outcome of the wargame.

The course-of-action selection stage is the decision and execution stage, in which the commander selects the highest-rated course of action, refines it to ensure clarity (commander's intent), and generates the plans and orders for unit execution.

Key Attributes of Doctrinally Correct Tactical Planning

Even from the above brief overview of tactical planning, it is possible to identify several key attributes of the overall doctrinally prescribed process:

  • Planning is doctrinalized and knowledge intensive.

  • Echelon affects the planning focus and process.

  • Planning strongly reflects the resource context.

  • Planning strongly reflects the task context.

Planning is Doctrinalized and Knowledge Intensive

Because of the complexity of the domain, the risk/reward ratios of the outcome, the limited resources available for finite-time plan generation, and, perhaps most important, individual differences in planning and decision making abilities, the planning process has been highly doctrinalized.3 As noted, step-by-step procedures are designed to take the planner from the current situation to the mission end state. This is true for all echelons, with the complexity of the planning process increasing with echelon (and time available). The planning process also relies on a detailed and extensive knowledge base of the domain,4 covering the environment, enemy capabilities, and friendly capabilities. At midlevel and higher echelons, the planning process is so knowledge intensive that its accomplishment demands a staff of specialists. It is perhaps as far from reasoning by first principles as one can imagine.

3  

This last point is brought out by Falleson (1993:9), who notes that the Prussian Army formalized the tactical decision-making process in the early 1800s so as "not to leave success to the rare chance of tactical genius."

4  

Provided in the U.S. Army by an extensive collection of field manuals, augmented by a broad array of training material and formal courses.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Echelon Affects the Planning Focus and Process

Although this review does not include instances of planning at different echelons, it is clear that the focus of the planning differs with echelon. At the lower echelons, much of the emphasis is on path planning and movement to an objective. For example, an individual may spend time planning a route that maximizes cover under military operations in urban terrain (MOUT), or, as in our vignette, a platoon commander will spend time planning a best route to an objective that his unit is to defend. At the middle echelons, say the company to brigade levels, route planning is clearly important, but greater emphasis is placed on coordinating the subordinate and collateral units in the operation, so that issues of synchronization, timing, and communication become more important. At still higher echelons, the division level and above, the same also holds, but now logistics planning takes on greater importance because of the longer time spans involved. It should be emphasized that we are not suggesting that lower-echelon planning excludes logistics concerns (since clearly the individual combatant is concerned with ammunition reserves) or that higher echelons are unconcerned with movement along possible avenues of approach. Rather, we are saying that the focus changes with echelon, long-term coordination and logistics issues taking on greater weight for the higher echelons.

Planning at different echelons is also characterized by different time windows afforded the operation and a corresponding change in the planning process itself. At lower levels, the operation and the planning for it may take from minutes to hours, as illustrated by our vignette. At higher levels, the operation may take days, and planning for it may take weeks to months. Because fewer resources are available at the lower echelons, this time compression serves to change not only the focus but also the nature of the planning process. Planning at the lower echelons becomes more reflexive and more driven by past experience in similar situations. Planning at the higher echelons is more contemplative and is supported by extensive rational analyses of the options.5 This suggests that different planning techniques and processes may be employed by the tactical decision maker at each echelon.

Planning at different echelons also affects how current situations (initial state), missions (final goal state), and actions (plan activities) are viewed by the planner. Planning and execution occur within a heavily hierarchical organization, so that one unit's planned tasks or activities become the plan goals or missions of that unit's subordinates. This hierarchical goal/task decomposition starts at the highest operational level, say the corps level, and proceeds down the organization to the lowest individual combatant. The result, in theory, is a com-

5  

Although the next section describes research that clearly disputes this viewpoint, the focus here is on "doctrinally correct" planning.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
mon planning process at each echelon, but one whose objects (unit situation, unit mission, and subordinate tasks/missions) change to reflect a unit's location in the hierarchy. In the artificial intelligence planning literature, this process is viewed as abstraction planning (Sacerdoti, 1974).
Planning Strongly Reflects the Resource Context

Planning is not conducted in an abstract world of perfect and complete information (as in, for example, a chess game), nor is it conducted as an open-ended exercise, free of time constraints. Effective planning relies on a proper interpretation of the mission plan goal specified by headquarters. As discussed at a recent Workshop on Commander's Intent (Anonymous, 1997), the goal may not be clear because of a failure to properly specify the intent, a failure in the communication channels, or a failure to properly interpret the intent. Effective planning also relies on an accurate assessment of the current situation. As discussed in Chapter 7, this accurate assessment depends, in turn, on an accurate perception of key tactical events or cues, which may be circumscribed and noisy and thus lead to erroneous planning decisions. Finally, the planning process itself is time constrained, so that the planning effort must be scaled down to the time available. This meta-planning skill of time management is illustrated in our vignette: one of the first of the platoon leader's activities after receiving the mission is to define a timeline for all activities leading up to occupation of the battle position, including planning, logistics preparation, and rehearsal. In this case, plan preparation takes somewhat less than 2 hours. In more dynamic situations, there may be considerably less time available, so that an abbreviated planning process occurs, with less extensive consideration of alternative courses of action, less wargaming, and so forth. ST 100-9 provides guidelines on how to speed up the decision making process under time stress.

Planning Strongly Reflects the Task Context

Planning is only one activity that occurs in the overall military operational context as part of the following sequence of activities:

  1. Reception of orders from the command unit, defining the plan goal and plan constraints (ST 100-9: mission analysis).

  2. Specification of the environment, the enemy situation, and the friendly situation (ST 100-9: intelligence preparation of the battlefield generation) and selection of a high-level plan for the unit and subordinate units (ST 100-9: course-of-action development, analysis, and decision).

  3. Communication of operational plans and operational orders to subordinate units for recursive plan elaboration (ST 100-9: decision and execution).

  4. Execution of plan (operations).

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
  1. Continued situation assessment to support monitoring of plan compliance through comparison of the assessed and the planned situation.

  2. Replanning whenever noncompliance exceeds a decision threshold or when an opportunity exists for improving on the planned outcome (opportunistic planning).

Note that only the first four steps are covered in detail in FM 101-5 (U.S. Army, 1997). The latter three steps, which actually occur in parallel and in the heat of battle, are clearly critical to the overall success of the operation. Being able to determine when a plan is failing and replan accordingly under adverse conditions and time stress would appear to be key to successful tactical decision making. From our review, however, it is not clear that this class of plan monitoring/modification activities is as well covered by doctrine as are the preoperations planning activities.

Key Attributes of Tactical Planning in Practice

To this point, the discussion has focused on tactical planning as prescribed by U.S. Army doctrine. If the ultimate objective of future efforts in human behavior representation is to model doctrinally correct planning behavior, this should be an adequate starting point for a requirements definition effort. If, however, future efforts in human behavior representation are to be aimed at modeling actual tactical planning, a serious knowledge engineering effort is needed to build a behavioral database describing planning during combat decision making. This conclusion follows from the observations of many researchers that actual planning behavior observed and reported differs markedly from that prescribed by doctrine.

Fallesen (1993) conducted an extensive review of the relevant literature in battlefield assessment and combat decision making over a 5-year span, covering hundreds of studies, reports, and research summaries. The interested reader is referred to this review for an in-depth discussion of the disconnect between doctrine and practice. The discussion here highlights some of the key points relevant to the planning process.

Management of the Planning Process

Overall management of the planning process and staff is often poor. There is inadequate involvement by the commander; Fallesen (1993:15) notes that this may be because "current doctrine establishes an unclear role for the commander." There is inadequate coordination across the staff specialties, with engineers and fire support officers (FSOs) often being left out. Finally, there is inadequate meta-planning over the scheduling of the planning activities and the allocation of staff resources. The net result is that the process defined by doctrine "typically is not followed closely" [emphasis added] (Fallesen, 1993:17).

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Information Exchange for Planning

The underlying information needed for good decisions is often not sought or not specifically related to the decision process. There is inadequate exchange of critical information across staff members, planners fail to seek actively the information they need (Thordsen et al., 1989), information generators fail to present interpretations or implications of the briefed information, and finalized plans are often not presented to the commander for final review (Metlay et al., 1985).

Situation Assessment to Support Planning Assumptions

Situation assessment for planning suffers from a number of shortcomings, including (Fallesen, 1993) failure to consider an adequate number of facts for accurate situation awareness, failure to verify assumptions made for planning purposes, failure to weight information according to its quality, failure to interpret information (Endsley's [1995] Level 2 situation awareness), and failure to make predictions (level 3 situation awareness). Although these shortcomings are discussed in detail in Chapter 7, they are noted here because of the critical importance of correct situation awareness for planning.

Course-of-Action Development

The creative heart of the planning process, course-of-action development, deviates along three key dimensions from what is doctrinally specified (Fallesen, 1993). First, the management and tracking of course-of-action alternatives, another meta-planning task, is poorly done, with no audit trails and inadequate documentation of assumptions and concepts. Second, course-of-action generation does not appear to follow the doctrinally prescribed model:

The tactical decision making model of the [doctrinally prescribed] process indicates that multiple options should be generated, and that options should be distinct from one another. Findings have shown that multiple options are often not generated, and that the options are not always unique. When three courses of action are produced, they are sometimes called the 'best', the 'look-alike', and the 'throw-away.' (Fallesen, 1993:24).

Fallesen goes on to say that this deviation from doctrine is not necessarily bad. He points out that Klein (1989, 1994) and Thordsen et al. (1989) term this behavior recognition-primed decision making (see Chapter 7) and believe it to be the normal operating mode of experienced decision makers:

[Thordsen et al. (1989)] concluded that multiple options were not considered as a matter of course nor were [the planners] compelled to conform to the traditional decision analytic model. Planners considered alternatives out of necessity if the first alternative proved infeasible (Fallesen, 1993:24).

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

This view clearly suggests a sufficing rather than an optimizing mode of plan generation. Finally, a third deviation from doctrine concerns the failure of planners to generate sufficiently detailed courses of action, including contingency plans for dealing with deviations from the plan.

Course-of-Action Analysis and Selection

The actual analysis and selection of courses of action also deviates significantly from doctrine. Doctrine prescribes extensive use of wargaming, concurrent evaluation of the different courses of action, and avoidance of premature decisions. What is observed, however, is a different matter (Fallesen, 1993). First, wargaming is often simply not conducted, or is done superficially or using noncomparable wargaming techniques (Fallesen and Michel, 1991). Second, course-of-action evaluation tends to be serial; that is, it is conducted depth first, rather than the doctrinally prescribed breadth first:

We found planners tended to employ a process where they would evaluate an option or idea by gradually examining deeper and deeper branches of the idea for workability. … If [an idea] is rejected the decision maker either moves on to a totally different option or idea or goes back up the deepening chain to a point (theoretically) above the source of the flaw and then follows another branch. (Thordsen et al., 1991:2)

In addition, the method for evaluating courses of action does not follow the recommended multiattribute decision matrix approach (Thordsen, 1989). Fallesen et al. (1992) explain:

Using a decision analytic approach, as complicated as a weighted, multiattribute utility matrix or as simple as a summary column of pluses and minuses, can be misleading for complex, dynamic tactical problems (Fallesen et al., 1992:87).

Finally, doctrine cautions against reaching early decisions about the desirability of one course of action over another. In practice, however, the course of action first generated is often selected (Geva, 1988), and it has been found that early decisions do not degrade performance in selection of the proper course of action (Fallesen et al., 1992):

Making early decisions about courses of action are contrary to a formal, analytical process, but making decisions early has been shown not to impact solutions adversely in one study and to promote better solutions in another (Fallesen, 1993:30).

As Lussier and Litavec (1992:17) point out, the time saved by making an early decision can be put to good use in detailed plan refinement, contingency planning, and rehearsal:

Other commanders said they must just bite the bullet and decide quickly. They emphasized that the important thing is how well planned and executed the mission is, not which course of action is chosen.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Plan Monitoring and Replanning

Once the operation is under way, ongoing battlefield monitoring is key to determining how closely the actual situation is following the plan. This situation awareness function, unfortunately, is not well carried out:

The 1992 analysis of NTC [National Training Center], JRTC [Joint Readiness Training Center], and CMTC [Combat Mamzuren Training Center] trends reported that 59 percent of brigade, battalion task forces, and company/teams did not track the battle quickly and accurately. The failure to do so created conditions for fratricide and being unaware of available combat power. (Fallesen, 1993:35)

This failure to track the battle also clearly created conditions for an inability to see when a plan was failing or, for that matter, succeeding beyond the staff's expectations. If adequate plan monitoring can be achieved, then plan failures can be dealt with quickly, provided that contingency plans have anticipated the failure involved. Otherwise, a replanning exercise must take place concurrently with operations. However, as Lussier and Litavec (1992:16) note:

Commanders distinguish two situations: limited time situations, with only a few hours of planning time available, and execution situations, where mission planning is occurring at the same time as execution. In the latter case, the changing tactical environment makes the doctrinal decision making process even less applicable. Most commanders believe that they are not given much doctrinal help in doing that truncation; each must develop his own techniques and planning processes.

Clearly, one can expect the real-time replanning effort to deviate significantly from the doctrinally specified preoperations planning process.

Implications for Modeling the Tactical Planning Process

The above discussion has a number of key implications for modeling the tactical planning process, related to the difference between theory and practice; the need for domain knowledge in model development; the need to model the various stages of planning; and the effects of echelon, resources, and context.

Theory vs. Practice

The preceding discussion makes a clear distinction between planning activities that are prescribed by doctrine and those observed in practice. Table 8.2 summarizes the implications in a matrix of procedural options (doctrinal, nondoctrinal) and execution6 options (perfect, imperfect) for the modeling of planning. The option selection by the modeler will clearly depend on the purpose for which the human behavior representation is to be applied. For example, in the

6  

By execution, we mean execution of the planning activity, not execution of the plan.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

TABLE 8.2 Options for Modeling Planning

 

Execution Options

 

Procedural Options

Perfect

Imperfect

Doctrinal

Doctrinally correct with error-free execution (blue subordinate HBR and/or red adversary HBR used for training novice commanders)

Doctrinally correct with error-prone execution (novice blue subordinate HBR used to train expert blue commander)

Nondoctrinal

Nondoctrinal with error-free execution (innovative red adversary HBR used to train expert blue commanders)

Nondoctrinal with error-prone execution (realistic blue subordinate HBR and/or red adversary HBR used for mission rehearsal)

NOTE: HBR, human behavior representations.

training area, one might take the upper left option (doctrinal, perfect) to represent blue subordinates during initial training of blue commanders; a transition to the upper right option (doctrinal, imperfect) might be made for advanced training in dealing with less competent subordinates, and a transition to the lower right option (nondoctrinal, imperfect) for commander evaluation under realistic conditions. Alternatively, the upper left option might be used to represent red adversaries during early blue force training, with a transition to the lower left option (nondoctrinal, perfect) for advanced training in dealing with innovative adversaries. Clearly, other options abound, especially if the scope is broadened beyond training to include systems evaluation, development of tactics, and the like.

Need for Domain Knowledge

Whether a doctrinal or nondoctrinal modeling approach to planning is taken, it is clear that any model will rely on a detailed and extensive knowledge base of the service-specific domain, covering the environment, friendly and enemy unit capabilities, operational constraints, general doctrine, and the like. It is evident that any computational planner must be capable of dealing with a highly complex domain, far removed from first-principle planners typifying early artificial intelligence approaches to planning.

Planning Stages

If a planning model is to generate planning behaviors that somehow mimic those of a human planner, the modeler must attempt to replicate the various stages of planning, especially if the goal is to achieve some measure of doctrinal fidelity. Specifically, a planning model 7 should attempt to represent the planning stages discussed in the previous section:

7  

That is, a model suited for the Army.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
  1. Information exchange/filtering/retrieval, a front-end active perception process, responsible for generating the scenario-specific knowledge base for assessment and decision making.

  2. Situation assessment, another front-end process responsible for generating informational abstractions and situational context to support the planning process.

  3. Course-of-action development, the process responsible for generating candidate plans. A decision-theoretic approach might be taken for doctrinally correct planning, whereas a recognition-primed decision making approach, relying on episodic memory, might be better suited to the modeling of actual planning behaviors. Whichever approach is taken, attention needs to be paid to whether an optimizing or a satisfying approach is used.

  4. Course-of-action analysis and selection, the process responsible for selecting the best among candidate plans. A decision-theoretic approach would provide a natural environment for incorporating multistage wargaming and evaluation activities. A recognition-primed decision making approach would bypass this stage.

  5. Monitoring and replanning, the processes responsible for assessing the situation and any deviations from the plan, and then developing or calling up new plans to compensate for those deviations.

Echelon Effects

The strong dependence of plan focus and process on echelon was noted earlier. Table 8.3 provides a qualitative overview of this echelon effect. It seems clear that any realistic planning model must account for this echelon dependence through the development of either separate echelon-specific models (as is the current practice) or a more generic planner that reflects echelon-dependent changes in goals, scope, staff resources, time planning windows, and the like. The latter approach may be feasible because of the heavily hierarchical command and control structures employed by the services, which lead to a natural hierarchical goal decomposition by echelon.

Resource Effects

Planning's strong dependence on the resources available, including information, time, and staff, was also noted earlier. Plans are based on incomplete, uncertain, and outdated information, and planning models need to reflect this fact. The best way of ensuring that they do so is to create a planning model that has no access to ''outside" information, but instead relies solely on perceived states and assessed situations, which, if modeled correctly, will realistically intro-

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

TABLE 8.3 Qualitative Trends in Planning Due to Echelon

Echelon

Focus

Process

High

Logistics, overall operation

Contemplative

Mid

Coordination, communication

(Not available)

Low

Route planning, cover

Reflexive

duce uncertainty and untimeliness.8 Plan generation is also time and resource (staff) constrained, and any realistic planning model must reflect this fact as well. One approach is to incorporate meta-planning capabilities in the model to budget time and allocate staff (subplanners) for the planning process, change planning strategies as a function of time/staff available,9 monitor plan generation progress, and the like. An added benefit of this approach is that the resulting planning model can begin to reflect the actual meta-planning activities of the human tactical decision maker, clearly adding to the face validity of the model.

Contextual Effects

Finally, it has also been noted that planning is only one activity that occurs in the overall operational context. It is not an isolated process, but depends highly on—and often competes for resources with—many other activities (e.g., information gathering, situation assessment). In modeling planning, it is therefore necessary to represent in-context planning, interruptions caused by higher-priority activities, attention allocation across concurrent tasks, and a variety of other issues that arise in multitask environments.

MODELS FOR PLANNING IN MILITARY HUMAN BEHAVIOR REPRESENTATIONS

The following three subsections (1) review planning models in existing military human behavior representations; (2) examine military decision-aiding systems that contain a planning component, since they may have potential use in the development of computational planning modules for future military human behavior representations; and (3) summarize the key findings in both areas. Table

8  

If some meta-knowledge of this incertainty/untimeliness is held by the planner, then it seems clear that a similiar meta-knowledge based should be maintained by a model. This meta-knowledge base would support modifications in plan goals, planning strategies, and the like as a function of the uncertainty of the planning knowledge base (i.e., high uncertainty should lead the planner to choose from robust, but possibly nonoptimal plans; low uncertainty should lead to the choice of more optimal plans, which may, however, be more likely to fail when the planning assumptions are violated).

9  

For example, using shortcuts under time stress.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

8.4 provides an overview of the human behavior representations, decision aids, and simulations reviewed in this section.

Planning Models in Military Human Behavior Representations

Adaptive Combat Model

The Marine Corps Modeling and Simulation Management Office has a program in Adaptive Combat Modeling, aimed at modeling individual and unit-level behaviors in three separate phases (Fisk, 1997):

  • Phase 1: Route Planning

  • Phase 2: Control of Movement

  • Phase 3: Terrain Exploitation

The objective is to develop adaptive autonomous synthetic agents for use in tutoring systems and command trainers, and possibly as a knowledge-base source in advanced decision aids for fire support. For Phase 1, route planning in the face of multiple constraints (e.g., terrain obstacles, time, fuel) is being modeled through the use of genetic algorithms that serve to generate, evaluate, and evolve quasioptimal route solutions satisfying the overall route constraints (risk, personal communication). To date, no effort has been made to match the agent's performance with that of a human route planner, although the possibility exists—if there is more of a focus on human behavior model development, as opposed to tactical decision-aid development.

Commander's Visual Reasoning Tool

A high-level descriptive model of the staff course-of-action planning process is presented by Barnes and Knapp (1997). Although not properly a model, it provides an overview (through a command/staff dependency matrix) of the course-of-action planning activities undertaken by the brigade commander and his intelligence officer (S2), operations officer (S3), and fire support officer (FSO). It thereby provides a framework for ensuring that all individual commander/staff functions are represented in any subsequent command and control (C2) human behavior representation development effort. It also identifies the key points of interaction among team members (e.g., "provide commander's guidance to staff," "coordinate with division G2"), thus identifying communication channels and information flow within and outside of the commander/staff team.

Barnes and Knapp (1997) also describe an experimental tool—the commander's visual reasoning tool (CoVRT)—for prototyping future brigade C2 environments by means of a three-workstation network supporting the battle management functions of the commander, S2, and S3. Such a research environment could be used for future model validation efforts in developing scenarios

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

and recording the detailed course-of-action planning process conducted by individual staff members. These records could then serve as a real-world database for evaluating the realism of comparable model-generated behaviors.

Dismounted Infantry Computer-Generated Force

The Hughes Research Laboratory has developed a computer-generated force (CGF) representation for the Marine Corps (see also Chapter 2), intended to model individual fire-team members, fire-team leaders, and squad leader to support the training of their respective superiors (Hoff, 1996). Individual behavior is simulated by a production rule system that generates a plan comprising a number of "task frames" that are eventually implemented by the modular semiautomated forces (ModSAF) dismounted infantry "action" module (see Chapter 3) operating within the distributed interactive simulation (DIS) environment.

The CGF plan generation module consists of 22 separate modules or rulesets, each containing 10 to 100 production rules, which encode a subject matter expert's ranking of decision alternatives for a number of key decisions, such as attack or abort, avoid contact, select assault point, and select route. For example, a select route decision rule might be:

IF (route A distance <50 m)

AND (route A terrain is slow-go)

AND (route B distance is 100 m)

AND (route B terrain is go)

THEN (select route B)

The distance measures are left "fuzzy" to allow for variations.

Ruleset antecedents are essentially features of the tactical situation (e.g., "enemy posture is dug-in"), which are obtained directly by querying the ModSAF module(s) responsible for maintaining the state of the simulation. Thus, no attempt is made to model the individual's cognitive functions of information gathering, perception, correlation, or situation assessment. Once the necessary state information has been gathered from the ModSAF modules, one frame of a plan is generated and sent to ModSAF for execution (e.g., "attack the objective in a wedge formation"). It is unclear how often the plan can be updated to account for unplanned circumstances, but the state-driven production rule approach appears to allow for almost continuous updating (e.g., "situated cognition").

Although components of the tactical planning function are segregated by ruleset, it is not clear whether there is any attempt at hierarchical planning (e.g., decide to attack first and then decide the route) and/or modeling of the interaction of interdependent or conflicting goals (e.g., attack quickly, but minimize casualties). Finally, it is unclear whether this is much more than a one-step-ahead planner, and thus effectively a purely reflexive agent, or if it can generate a phased sequence of frames defining the overall evolution of the plan from start to finish.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

TABLE 8.4 Models for Planning in Military Human Behavior Representations (HBR)

Name

Sponsor

Service (Domain)

Level

Functions/Activities Modeled

Implementation/Architecture

Comments

Adaptive Combat Model

Marine Corps Modeling and Simulation Office

Infantry/Marines (HBR models)

Individual and unit

Route planning

• Genetic algorithm optimization

• Currently an exploratory study

Commander's Visual Reasoning Tool (CoVRT)

Army Research Laboratory (ARL)

Army (decision support)

Brigade

Course-of-action generation

• Decision-aiding workstation with graphic depiction of military entities

• Underlying command/staff dependency matrix provides over-all framework for information flow among HBR agents

Marine Computer-Generated Force (CGF)

Defense Advanced Research Projects Agency (DARPA), Hughes Research

Marine (HBR models)

Individual and unit

Route planning

• Production rule system driven by external state variables

• Multiple rulesets for different decision alternatives;

• Generation of single plan frame for ModSAF execution

Computer-Controlled Hostiles for Team Target Engagement Simulator (TTES)

Institute for Simulation and Training/University of Central Florida (IST/UCE)

Marine (HBR models)

Individual

Short-term planning for individual military operations in urban terrain activities

• Hierarchical goal decomposition;

• Decision-theoretic goal selection;

• Situation-driven rules

Appears to be single-step planner, but could be expanded to multiple steps for individual course-of-action generation

Fixed-wing Attack (FWA)-Soar and Soar-Intelligent Force Agents (IFOR)

DARPA, Air Force, University of Michigan (UMich), University of California/Los Angeles (UCLA), Carnegie Mellon University

Air Force (HBR models)

Individual and unit

Full activities of tactical pilots across a wide range of aircraft and missions

• Soar architecture with hierarchical goal decomposition;

• Efficient production rule system to deal with large rulebase;

• Situation-driven rules

• Planning not explicitly represented, as Soar supports only single-step planning;

• Could be expanded to support an explicit planning module

Rotary-Wing Attack (RWA)-Soar

DARPA, Army, UMich, UCLA

Army Aviation (HBR models)

Individual and unit (company)

Full activities of rotorcraft pilots and company commander for RWA mission

• Live battalion commander;

• Soar-CFOR company commander that:;—Generates mission plan;—Monitors progress;—Replans as necessary;

• Soar-IFOR RWA pilots;

• ModSAF vehicle entities

• Plan generation and elaboration done through tactical templates and standard operating procedures;

• Plan refinement done through checks on task interdependencies, timing

Man-Machine Integration Design and Analysis System (MIDAS)

National Aeronautics and Space Administration, Army

Army Aviation (HBR models)

Individual and unit

Full activities of tactical rotorcraft pilots

• Symbolic operator model architecture with hierarchical mission activity decomposition;

• Production rule system to implement procedures;

• Situation-driven productions

• Similar to Soar in its top-down decomposition from mission phases to low-level activities;

• Single-step planning, but could be expanded to support an explicit planning module

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
This page in the original is blank.
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

Name

Sponsor

Service (Domain)

Level

Functions/Activities Modeled

Implementation/Architecture

Comments

Naval Simulation System (NSS)

Navy

Navy (HBR models)

Command and control (C2) and unit

Full activities of a Navy task force at entity level

• Decision tables or prioritized production rules;

• Future state predictor that supports some planning

• Future state predictor could be used to support reactive

Automated Mission Planner (AMP)

IST/UCF, University of Florida

Army (HBR models)

Unit, company

Course-of-action generation

Four stage process:;

1. Terrain analysis for route planning;

2. Course-of-action generation;

3. Course-of-action simulation;

4. Course-of-action selection

• For use in ModSAF company commander entities;

• Development status unknown;

• Key use of simulation-based plan evaluation

 

International Advisory Group-European Cooperation for the Long Term In Defense Consortium (ISAG EUCLID)

Army and Navy (decision support)

Unit

Course-of-action generation, maneuver and FS planning. AGW planning

• Multiagent architecture

• Broad Europe-wide effort in decision-support tools;

• Planning modules may be useful in HBR modeling

Battlefield Reasoning System (BRS)

ARL Federated Laboratory, University of Illinois at Urbana-Champaign

Army (decision support)

C2

Course-of-action generation

• Blackboard architecture with multiple specialist agents;

• Current focus on course-of-action generation

• ARL-sponsored effort in decision-support tools;

• Course-of-action planning modules may be useful in HBR models

Decision Support Display (DSD)

ARL Federated Laboratory, North Carolina Agricultural and Technical State University

Army (decision support)

C2 Unit

Course-of-action generation and logistics planning

• Multiple distributed rulesets

• Currently an exploratory study;

• Rulesets may be useful for future model development

Computer-Controlled Hostiles for SUTT

The small unit tactical trainer (SUTT)10 is used to train Marine rifle squads in MOUT clearing (Reece, 1996; see also Chapter 2). The virtual world is populated by computer-controlled hostiles, which have been developed to behave as "smart" adversaries that evade and counterattack blue forces. The key state variables are fairly low level (e.g., soldier position, heading, posture), and perception and situation awareness are based on simple detection and identification of enemy forces.

Most of the activity appears to be fairly reflexive, but there is an attempt at situation-driven planning using hierarchical goal decomposition. High-level goals

10  

Formerly known as the team target engagement simulator (TTES).

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

are decomposed into subgoals, which can be either concurrent or alternative. For example, a concurrent goal set for attack and move is to attack and move simultaneously; an alternative goal set, say, for the attack subgoal, is to perform a quick fire or aimed fire. For each high-level goal, a full tree of candidate goals and subgoals is constructed, down to the leaf action nodes, prior to any action generation. Deconflicting of conflicting concurrent activities and selection of nonconflicting alternative activities are performed by calculating the effective utility of each path in the goal tree in a fashion similar to the usual decision-theoretic approach based on priority values assigned to each goal/action node (see Chapter 6). The action associated with the maximum utility goal path is then selected, and this action specifies the entity's behavior in the next time frame.

This type of hierarchical goal decomposition approach to action selection could subserve a tactical planning function, but in the current implementation it

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

appears that only a single action is selected at any one time frame and not a sequence of actions, as one would expect in any sort of plan that spanned a finite time window (from current situation to future goal). What has been implemented is, in effect, a situation-driven decision-theoretic reflexive action generator, operating over a planning horizon of a single time frame. It may be possible to convert the current planner into a full-fledged multistep plan generator, although it is unclear how this would be achieved in the current implementation.

Fixed-Wing Attack (FWA)-Soar and Soar-Intelligent Forces (IFOR)

Considerable effort has been devoted to applying the Soar cognitive architecture (Laird et al., 1987; Laird and Rosenbloom, 1990) to the modeling of human behavior in fixed-wing tactical missions across three services: the Air Force, the Navy, and the Marines (see Chapter 2). Initial efforts have led to a limited-scope demonstration of feasibility for FWA-Soar (Tambe et al., 1995); more recently, FWA-Soar was used in the synthetic theater of war (STOW)-97 large-scale warfighting simulation (Laird, 1996).

The standard Soar approach to hierarchical goal decomposition is taken, so that a high-level goal (e.g., "conduct intercept") is broken down into its component subgoals (e.g., "adjust radar," "achieve proximity"). Each of these subgoals is then successively broken down until some implicit ''atomic-level" goal is reached (e.g., "press trigger"), beyond which no further subgoals or activities are generated. The successful achievement of any given goal in the hierarchy typically requires the satisfaction of one or more associated subgoals, so that a tree structure naturally arises. Thus in the tactical air domain, one might see the following tree structure (modified from Laird, 1996):

Intercept

Search

Select threat

Achieve proximity

Employ weapons

Select missile

Get threat in weapons employment zone (WEZ)

Launch missile

Get steering circle

Get lock-on

Push launch button

Do FPOLE maneuver

In the above, indenting is used to indicate the level in the tree, and at each level, only one node (shown in bold) is expanded. Satisfaction of any given goal may require the satisfaction of all its subgoals (for example, the launch missile goal shown above) or of only some subset of its subgoals (for example, perform

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

evasive turn could require a left or a right turn). In addition, subgoal sequencing may be specified by a need to satisfy antecedent goal conditions (e.g., in the launch missile sequence, push launch button must be preceded by achieving lock-on). Finally, arbitration or deconflicting of concurrent conflicting goals (e.g., evade threat 1 and attack threat 2) is accomplished by assigning fuzzy preferences (e.g., worst, poor, neutral, good, best) and, on the basis of these symbolic preferences, selecting the most preferable.

Implementation is accomplished by means of a highly efficient production rule system, with all triggered rules being fired until quiescence, followed by a deconflicting/prioritization stage for selecting the behavior(s) in which to engage for that particular time frame (nominally every 50 milliseconds, but frame times can be less frequent with more complex scenarios). In FWA-Soar, approximately 400 goals are represented, with about 4,000 rules specified on how to achieve them (Laird, 1996).

In generating a path through the goal tree and subsequently pruning off lower-priority paths, Soar has the capability of generating a linear sequence of goals and subgoals to be followed over some finite time horizon, starting with the current time frame. This linear sequence of goals and their associated behaviors is effectively a plan: it is a linear sequence of behaviors that will transform the current state of the system into some desired goal state through a path of sequential system states defined by the subgoals along the way. It is an open-loop plan in that it has been generated for the current state of the system and the current goal; any uncontemplated change in either, or any uncontemplated failure to reach a particular subgoal in the path, may cause the plan to fail. However, this is not a problem in the Soar implementation since replanning occurs every time frame, so that plan deviations can be detected and accounted for via a (possibly) different path through the goal space. By replanning every time frame, Soar effectively converts an open-loop plan into a closed-loop plan.

Essentially, then, Soar acts as a purely reflexive agent11 in the cited Air Force applications, with no clearly specified planning function (although the latter could be implemented through the subgoal path training just noted) (see also Tambe et al., 1995). Clearly, additional development is needed, and a likely candidate for that development would appear to be the subgoal path-tracing approach outlined here. Implementation of this approach would require: (1) the specification of an explicit planning task, which would generate the open-loop subgoal path for some finite time window into the future; (2) the storage of that plan in memory;

11  

Tambe et al. (1995) distinguish Soar from pure “situated-action” systems (Agre and Chapman, 1987) in that the latter reflexive systems are typically driven by external world states (e.g., actual threat position), whereas Soar is driven by internal estimates of those states (e.g., perceived threat position). Although the distinction can lead to significantly different behavioral outcomes, it is irrelevant with regard to planning functionality: neither incorporates planners, and both are thus purely reflexive.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

(3) the generation of appropriately time- or event-driven behaviors selected to achieve each subgoal in sequence; and (4) the monitoring of conformance with the plan in memory. Nonconformance would trigger a replanning function, starting the process over again as a reflexive function driven, say, by inadequate planning, unanticipated changes in the world state, and the like. In this fashion, Soar could balance a purely closed-loop reflexive production rule approach to behavior generation with a purely open-loop plan-based and scripted approach to behavior generation. The former approach clearly imposes greater workload per computation cycle (with full replanning every frame), but is much more responsive to unanticipated changes in world state; the latter is a lower-workload solution with replanning occurring only as required, but is less responsive to plan failures or unanticipated world state changes.

Rotary Wing Attack (RWA)-Soar

Soar has also been used to model an Army rotary-wing attack (RWA) mission, specifically an attack aviation company commander and subordinate pilots flying the attack helicopters (Gratch et al., 1996; Gratch, 1996; see also Chapter 2). A layered architecture is used, consisting from the top down of the following:

  1. A "live" battalion commander, specifying the overall mission by means of the structured command language CCSIL, communicating with:

  2. a simulated company commander modeled as a Soar-command forces (CFOR)12 entity, which receives the battalion order and generates a detailed mission plan, and then communicates the detailed operations order, again in CCSIL, to each:

  3. RWA pilot, modeled as a Soar-IFOR13 entity that receives individual task orders and generates the appropriate entity behavioral goals (e.g., vehicle flight path parameters, weapon employment activities) to command:

  4. the corresponding ModSAF entity (the pilot's rotorcraft) to pursue the mission plan within the DIS operating environment.

Naturally, there is also an upward information flow of hierarchically aggregated status information from the DIS environment, to the ModSAF entity, to its Soar-IFOR controller (the pilot), to the Soar-CFOR commander (the company commander), and finally to the live battalion commander.

12  

A further description of the CFOR program is provided by Hartzog and Salisbury (1996) and Calder et al (1996); see also Chapter 2. The planning component uses an approach based on constraint satisfaction to link together component activities, which are doctrinally specified, and which, when fully linked, form an activity plan, taking the command entity from the current situation to the desired end state. Specific details of how activities are linked efficiently to avoid a combinatorial explosion of options through the planning space are described in calder et al. (1966).

13  

See the earlier description of the Soar-IFOR program for fixed wing pilot modeling.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

From the planning point of view, the key component of the RWA-Soar architecture is the mission planning function executed by the Soar-CFOR representation of the company commander. The basic process involves three stages:

  1. Obtain the CCSIL battalion order, which specifies (a) the tactical situation, enemy and friendly; (b) the mission and the sequence of tasks to be executed; (c) the general means of execution; and (d) any coordination requirements.

  2. Generate the detailed company order for the individual RWA entities, using (a) prestored templates (actually rules) for elaborating the battalion orders, and (b) standard operating procedures (SOPs) or rules for filling in unspecified components of the battalion orders.

  3. Refine the company order by looking for inconsistencies and ensuring that intertask dependencies and timing constraints are met.

Note that the second stage of this planning process—essentially open-loop plan generation/elaboration—incorporates some general planning capabilities along the lines originally proposed by the general problem solver (GPS) of Newell and Simon (1963), in particular, step addition when necessary to accomplish the preconditions for an individual in an overall plan. As noted by Gratch (1996), Soar-CFOR adopted a plan representation approach based on hierarchical task networks (Sacerdoti, 1974; Tate, 1977; Wilkins, 1988)—a graphical means of hierarchical task decomposition with a specification of preconditions required for individual task execution. Plan modification occurs through refinement search (Kambhampati, 1997), which adapts plan segments to yield a complete sequence of steps, taking the planner from the initial to the goal state. Replanning occurs in the same fashion as does initial plan generation, so that dynamic unanticipated changes in the battlespace can be accounted for.

Finally, although the above discussion focuses on plan generation/elaboration, it should be noted that RWA-Soar also performs the key functions of plan monitoring and replanning. Plan monitoring occurs during execution of the plan and requires a comparison of the (perceived) tactical situation with the tactical situation anticipated by the plan. If the two do not match sufficiently closely, replanning is triggered. A production rule framework is used to check for antecedent satisfaction for plan monitoring and to maintain a memory of the plan goal for subsequent backward chaining and replanning.

Man-Machine Integration Design and Analysis System (MIDAS)

MIDAS is an agent-based operator model used in helicopter crew station design and procedure analysis (Banda et al., 1991; Smith et al., 1996; see also Chapters 3 and 7). A considerable amount of work has gone into developing a number of rule-based cognitive models, each specific to a particular design case and mission specification.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

Development of these cognitive models begins with the specification of a nominal scenario and then proceeds through a process of mission decomposition akin to the hierarchical goal decomposition performed in applying Soar to any realistic modeling task. With MIDAS, one would typically decompose a flight mission into mission phases, say, pretaxi, taxi, takeoff, …. The pretaxi phase would be broken down further into Soar-like procedures or goals, as follows:

Pretaxi

Preflight check

Takeoff clearance

Engine start check

Exhaust gas temperature check

Oil pressure check

Oil temperature check

Engine start light check

Subsystems check

Release brakes

As with Soar, the above shows the hierarchical nature of the procedure structure and how some procedures (in bold) can be broken down into subprocedures. Also as with Soar, a production rule system is used to represent these procedures, with the antecedents chosen to ensure proper sequencing relations and priorities. There are other similarities between MIDAS and Soar (specifically FWA-Soar and Soar-IFOR), such as relying on internal perceptions of external states, but it is this common approach to goal/procedure decomposition that suggests, at least in the planning realm, that MIDAS and Soar are effectively equivalent representational approaches: Both are reflexive production rule systems,14 in which the production rules representing behaviors at various levels of abstraction are carefully crafted to generate the appropriate behavioral time/event sequences for the scenarios and environment for which they were designed. Both are one-step or single-frame planners, but, as discussed earlier for Soar, both could be transitioned to multiframe planners, apparently without great difficulty.

Naval Simulation System

The naval simulation system (NSS) was developed to support the simulation of Navy operations in support of tactical analyses, decision-support applications, and training (Stevens and Parish, 1996; see also Chapter 2). The NSS represents tactics through what is termed a decision table, although it might more properly be termed a prioritized production rule system. A typical rule looks like the following:

14  

Again, reflexive with respect to internal or perceived world states.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

IF (situation x is true)

AND (event y is true within criteria z)

THEN (do action a with priority p)

The situational antecedent shown in the first line above ensures that only situationally relevant (e.g., for the right mission, in the right phase of the mission) rulesets are considered. The event trigger shown in the second line provides a "fuzzy" Boolean check by allowing for variable closeness in matching, through the parameter Z. Finally, the consequent in the third line specifies the action and also assigns it a priority, so that deconflicting or scheduling can occur when multiple actions are triggered in a given frame.

Operational plans in the NSS are represented either as open-loop plans, scripted by the user, or as reflexive, generated in the fashion just described by applying the appropriate tactical ruleset to the current tactical situation. Thus instead of "do action a with priority p," we might see "generate plan a with priority p." However, it is unclear whether actual plan generation is triggered by this production rule, or prestored contingency plans are retrieved from memory and issued by the command entity. An intermediate possibility, however, is suggested by the existence of another NSS module, the future enemy state predictor. Its function is to project the future battle status of the enemy some finite time into the future. Given this predicted future tactical situation, one could then apply the appropriate future tactical response. Since this situation would be in the future, it would become a response goal. Repetitive applications at multiple future time frames could then be used to generate multiple response goals, in effect producing a sequential plan for future actions and goals. Thus there would appear to be a mechanism (implemented or not) within the NSS for generating a tactical plan based on the anticipated evolution of the current tactical situation.

Automated Mission Planner

The University of Florida, in cooperation with the Institute for Simulation and Training at the University of Central Florida, is developing an automated mission planning (AMP) module for eventual incorporation into a command entity representing a company commander, as part of a ModSAF development effort within the DIS community (Lee and Fishwick, 1994; Karr, 1996). A four-stage planning process is outlined:

  1. A terrain analyzer generates possible routes based on the current situation, the mission orders, and the terrain database.

  2. A course-of-action generator uses "skeletal plans" (i.e., plan templates) to generate candidate courses of action identifying specific unit roles, routes, and tactical positions.

  3. A course-of-action simulation runs through each candidate courseof

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

action, evaluating it in terms of critical factors (e.g., how long a unit is exposed to enemy fire).

  1. A course-of-action selector selects the best course of action based on the scores from the evaluation.

AMP would appear to have many functions in common with the battlefield reasoning system (BRS) discussed below and currently being developed as a battlefield decision aid.

A key feature of the AMP approach is the introduction of simulation-based planning, wherein generated plans are evaluated through fast-time simulation in order to predict expected outcomes and provide a basis for ranking candidate courses of action according to their anticipated outcomes. Note the similarity to the doctrinally specified wargaming stage of course-of-action analysis, in which commanders conduct mental simulations or verbal walkthroughs of candidate courses of action to anticipate problems and evaluate outcomes. It would appear that this type of simulation-based evaluation should be a key component of any planning model considered for human behavior representation.

A concurrent effort at the Institute for Simulation and Training is the development of a unit route planning module for modeling route planning at multiple echelons (battalion, company, platoon) (Karr, 1996). The effort is currently focused on route finding using an Astar-search algorithm designed to find the optimal path based on a cost function defined over path length, trafficability, and covertness.

Planning Components of Military Decision Aids

International Simulation Advisory Group (ISAG) Effort in Planning Decision Aids

ISAG, a European consortium aimed at demonstrating how artificial intelligence and human-computer interaction tools can be applied to command, control, communications, and intelligence (C3I) systems development, has efforts under way in three areas (Ryan, 1997):

  • Automated report analysis

  • Army decision support, planning, and tasking

    • Terrain analysis

    • Course-of-action development

    • Maneuver and fire support planning

  • Navy decision support, planning, and tasking

    • Tactical threat evaluation

    • Antisubmarine warfare planning

    • Engagement coordination

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

The objective is to develop decision aids, not human behavior representations per se, but the overall approach is based on a multiagent architecture in which "intelligent" autonomous agents specialize in particular subtasks and coordinate their efforts through interagent communication protocols. Development appears to be in the early stages, but several of the focus areas may hold promise for the development of agent-based planning models for human behavior representation, particularly in Army course-of-action development, Army maneuver and fire support planning, and Navy antisubmarine warfare planning.

Battlefield Reasoning System

The battlefield reasoning system (BRS) is currently under development by the Beckman Institute of the University of Illinois, Urbana-Champaign, as part of the Army Research Laboratory's Federated Laboratory in Interactive Displays. The BRS is an overall architecture designed as a decision aid to support a number of reasoning tasks occurring in battlefield management, including (Fiebig et al., 1997):

  • Terrain analysis

  • Course-of-action generation

  • Wargaming

  • Collection planning

  • Intelligence analysis

The architecture is built around five agents, each specializing in one of the above tasks, a common data bus for interagent communication, a number of shared blackboards (e.g., offensive and defensive courses of action), and a number of shared databases (e.g., map data). The BRS is intended to serve as a prototype decision aid at the battalion through corps echelons, but two aspects of the system suggest that it may have potential for the development of future models for human behavior representation: (1) the development of each task agent is founded on the protocol analysis of a subject matter expert solving the same task, so that, at least on the surface, the agent should follow a reasoning process that somewhat mirrors that of the human decision maker; and (2) the agent-oriented architecture itself may serve as a good (normative) model of how to compartmentalize the multiple concurrent tasks facing the battlefield decision maker, providing a basis for subsequent matching of observed human behavior in comparable multitasking situations.

The current focus of BRS development is on the course-of-action generation agent, which reduced to its essentials, operates in accordance with the following process:

  1. Obtain the output of the terrain analysis agent, showing the go/no-go regions and other salient terrain attributes within the area of interest.

  2. Generate a modified combined obstacles overlay that populates the ter-

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

rain map with the red and blue forces, the attack objectives, and the potential avenues of approach.

  1. Specify the commander's initial guidance, which essentially identifies the mission, defines key subobjectives, and identifies key constraints to be met while accomplishing the mission (e.g., "use armored units only in reserve").

  2. Generate a candidate course of action by matching some combination of subordinate units to some combination of avenues of approach. Filter out any combination that fails to meet the commander's initial guidance specified in the previous step. Methodically cycle through all possible combinations (i.e., generate all possible courses of action) for the given set of units and avenues of approach.

  3. Present the filtered (i.e., commander-acceptable) courses of action to the G2 and G3 planners for further evaluation and narrowing down using the wargaming and collection agents.

As currently implemented, the operational planning component of the BRS focuses on course-of-action generation through a simple combinatorial generation of potential unit/avenue-of-approach matches, followed by subsequent filtering out of courses of action that do not comply with the commander's initial guidance. Clearly other planning factors need to be taken into account (e.g., fire support, logistical support), as well as other attributes of plan "goodness" beyond compliance with the commander's initial guidance (e.g., robustness to uncertainty). In addition, realistic courses of action need to account for time-phased activities of both blue and red units (e.g., through a synchronization matrix), so that considerable branching can occur even in one "baseline" course of action. Whether the exhaustive combinatorial approach now implemented in the BRS can handle the true complexity involved in the course-of-action planning task is not clear at this point. If so, it still remains to be seen whether this approach maintains some degree of psychological validity with respect to what is seen with the operations staff.

Decision Support Display

The effort to develop the decision support display (DSD) is described by Ntuen et al. (1997), who are members of the Federated Laboratory in Interactive Displays. The DSD is designed to support the commander and his/her staff as they deal with information at different steps during the following four-stage process:

  • Perception

  • Situation awareness and recognition

  • Situation analysis

  • Decision making

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

No explicit attempt has been made to model the commander's planning function, although the system codifies a number of rules covering components of course-of-action development, such as rules for resource use, rules for combat readiness, rules for battle engagement, and so on. It is unclear how the DSD will use these component rules to aid the commander, and in particular how they can be used for development of a cohesive battle plan over space and time. However, the DSD knowledge base may be useful in other applications, particularly planning models for human behavior representation, if the rulebase effectively codifies current tactical guidelines, rules, and decision heuristics. As this is an ongoing project under the auspices of the Federated Laboratory program, it should be rereviewed again at a later date.

Summary

For the set of models reviewed, the following four general approaches to model development have been taken.

Production Rule or Decision Table. The most common approach relies on production rules or decision tables. The rules/tables often follow doctrine closely, thus yielding greater face validity, at least for doctrinally correct behavior. Rules can be generated hierarchically to support a goal decomposition (Soar) or mission decomposition (MIDAS) hierarchy. The approach suffers the usual problem of brittleness, although fuzzy logic may offer a way around this problem. Also, the approach is unlikely to generate novel or emergent behavior, especially in situations not well covered by the rulebase.

Combinational Search or Genetic Algorithm. Used in some planning decision aids, this approach can be used for exhaustive generation of option sequences (plans) when the option spaces are well defined. Rapid course-of-action generation followed by constraint-based filtering of undesirable courses of action is one technique for developing sufficing plans. A related approach is to use genetic algorithms to generate plans through replication, mutation, and selection based on a plan "fitness function." Experience with other genetic algorithm applications suggests that this approach may prove quite useful in developing novel plans for novel situations, but it is unclear how difficult it will be to enforce doctrinally correct behavior when called for. In addition, a genetic algorithm approach may be quite computationally expensive in complex planning domains, precluding its use in real-time human behavior representations.

Planning Templates or Case-Based Reasoning. Planning templates have been used to expand mission orders into more detailed plans in accordance with well-specified doctrine. SOPs can fill in mission plan details, again in accordance with doctrine. These details can then become mission orders for subordi-

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

nate units, thus supporting a hierarchical application of the approach at each echelon of human behavior representation. Although no specific examples of a comparable approach based on case-based reasoning were reviewed here, it would appear that a case-based reasoning planning model could also have considerable potential because of both its capabilities for encoding extensive domain knowledge and its ability to model the experienced commander's apparent reliance on episodic memory for rapid plan development.

Simulation-Based Planning. Although not a purely generative technique, simulation-based planning relies on fast-time simulation of candidate plans to support rapid evaluation, modification, elaboration, and optimization of plans. Our review indicates that this is not a popular approach, either as a decision aid or as a planner for human behavior representation, but it may show more promise as better simulation models are developed and computational costs decrease. Simulation-based planning could also be used more in a final evaluative mode to model doctrinally specified wargaming of candidate courses of action.

The review of military planning models in this section reinforces a key finding of the previous section on tactical decision making: military planning is extremely knowledge intensive. Although early artificial intelligence planners emphasized general problem solving capabilities in an effort to develop domain-independent ''planning engines," it seems clear that the military planning domain, especially Army planning, requires that planning be conducted in accordance with an extensive doctrine covering operations and intelligence at all echelons in the military hierarchy. Thus, the method by which planning is to be conducted is specified, and it seems highly unlikely that any approach based on generic algorithms or artificial intelligence would converge on the existing set of planning techniques embraced by the military.

The result is a need for a highly knowledge-based approach to the modeling of existing military planning functions, particularly the process of transforming incoming operations orders into detailed plans for subordinate units. Thus, it would appear that wherever planning functions are specified explicitly by doctrine, a production rule or decision table approach could be used effectively to represent the process, especially at the lower unit levels, where course-of-action activities appear to be much more explicitly characterized. The same may also be true at higher unit levels, but here we suspect that a template or case-based approach may be more effective because of the additional complexity involved at these levels and the limited potential for solving complex problems with a simple rule-based system. In any case, it seems clear that to model military planning behavior, much of which is doctrinally specified (by either rules or cases), it is necessary to model the process with methods that explicitly embed that doctrine. A general-purpose planner would seem doomed to failure in this effort.

The military planning models reviewed here also conform to our earlier

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

findings regarding echelon effects. At lower levels (individual to company), planning focuses on path planning and movement. It may be that more algorithmic approaches are appropriate here (e.g., genetic algorithm optimization, dynamic programming) to solve essentially skill-level problems. At middle levels (company to brigade), path planning and unit movement are still key, but so are communication and coordination. Perhaps approaches that better represent explicit formalized doctrine are better here (e.g., production rule or case-based approaches) to solve what appear to be more rule-level problems. Finally, at higher levels (division and above), more emphasis is needed on conceptualizing the overall operation and specifying the mission. Whether the latter are best modeled by a higher-level decision-theoretic approach, say, or a case-based approach drawing on thousands of years of military history, is unclear at this point. It should also be recognized that the military hierarchy and the command and control information flow within that hierarchy (i.e., operational orders to subordinates and situational status from subordinates) are an ideal match for the development of hierarchical human behavior representation planners distributed across echelons. Thus at high echelons, only highly abstracted plans are formulated. As these plans are distributed downward, each subordinate unit elaborates on them (possibly through templates or production rules) and redistributes them downward as a new set of orders. Thus, plan granularity grows increasingly finer as one progresses downward in echelon, much as it does in a number of abstraction planners in the artificial intelligence community (e.g., Abstraction-Stanford Research Institute Planning System [ABSTRIPS]) or goal/task decomposers in the human behavior representation community (e.g., Soar or MIDAS).

The review of models in this section also reinforces our earlier findings about how planning relies critically on perception and situation assessment. The planning function must be given not only a goal, but also a specification of the starting point. That is, it must be given a description of the current situation and a specification of any relevant problem variables that might affect the planning solution. This is the function of "upstream" perceptual and situation assessment submodels, which are represented at varying levels of fidelity in the models reviewed here:

  • In some cases, neither the perceptual nor situation assessment functions are explicitly represented (e.g., FWA-Soar), so that environmental state variables (e.g., the identity and location of a bogey) are directly available in an error-free fashion to support either proactive planning or reactive behavior. This is clearly a low-fidelity representation of the sensors, the communications links, and the human perceptual apparatus that all serve to process and distort the information upon which plans are based.

  • In other cases, the perceptual functions are represented at some level, but no attempt is made to model higher-level situation assessment activities (e.g., computer-controlled hostiles for TTES). Thus one might model line-of sight

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

limitations in "seeing" a potential threat, but not model the abstraction of "seen" information (such as recognizing an attack formation that needs to be defended against). Unfortunately, by modeling only low-level perceived events (e.g., threat present at location A), one can generate only a low-level description of the current situation (e.g., three threats at locations A, B, C). Thus any subsequent planning can take place only at low levels, since the specification of the current situation is not sufficiently abstracted.

  • In yet other cases, both the perceptual and situation assessment functions are represented, so that low-level events are incorporated (in an appropriately errorful manner depending on sensor, communications, and perceptual limitations), as well as higher-level situations. This situation assessment function (e.g., recognizing that the three threats at locations A, B, C are in a defensive posture and unlikely to attack in the near future) then allows for situationally relevant rather than event-driven planning (e.g., select a course of action that is best for a dug-in threat group). This capability supports top-down hierarchical planning, or planning in context, clearly something in which most expert military planners engage.

Finally, the review in this section reinforces the requirement to regard the planning process as only one activity that occurs in the overall military operational context. The models reviewed here address, to one degree or another, the need to represent several stages of planning, starting with receipt of orders and proceeding to situation assessment of enemy and friendly dispositions, generation/evaluation/selection of a plan, communication of the plan to subordinates, plan execution, plan monitoring, and finally replanning. Modeling approaches for dealing with multitasking of this sort are discussed further in Chapter 4, and the findings presented there are clearly appropriate to the modeling of in-context military planning.

PLANNING MODELS IN THE ARTIFICIAL INTELLIGENCE AND BEHAVIORAL SCIENCE COMMUNITIES

An extensive literature on planning models exists in the artificial and behavioral science communities, and it is beyond the scope of this study to do much more than provide the reader with pointers to the relevant literature.

Excellent although slightly out-of-date reviews of artificial planners are provided by Georgeff (1987) and by Tate et al. (1990). They describe the basic chronology, starting with general problem solver (GPS) means-end analysis (Newell and Simon, 1963) and proceeding through a series of ever more sophisticated planners operating in increasingly complex domains. Starting with the essential definition of a plan as a sequence of operators transforming a problem's initial state into a goal state, Tate et al. (1990) outline the essential characteristics of artificial intelligence planners as follows.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

Planning is a search problem. Generating an operator sequence is a search in operator/state space, so search algorithms/heuristics are paramount. Tate et al. (1990) review and critique a dozen techniques, including Newell and Simon's (1963) early work with GPS as a formalization of means-end analysis, Sacerdoti's (1975) work with procedural nets in nets of action hierarchies (NOAH), and Tate's (1977) work with partial plans in the nonlinear planner (NONLIN).

Plan goals/tasks can be hierarchized. Goals have subgoals, and some goals can be independent of others. The way goals are organized is also key to planning. Tate et al. (1990) review a number of techniques, including the hierarchical planner ABSTRIPS by Sacerdoti (1974); work by Waldinger (1977) that deals with multiple simultaneous goal task satisfaction, and the system for interactive planning and execution monitoring system (SIPE) by Wilkins (1984); which supports the generation of plans that are both hierarchical and partially ordered.

Plans can be partial and nonlinear. Some planners generate a linear sequence of operators; others generate "partial-order" sequences, with segments to be joined later for a complete plan. There are almost a dozen ways to do this, starting with the goal interference checks conducted by Stanford Research Institute problem solver (STRIPS) (Fikes and Nilsson, 1971), and proceeding to more sophisticated interaction corrections accomplished by NOAH (Sacerdoti, 1975) and NONLIN (Tate, 1977).

Plan goals/tasks can have conditionals. Plans can have conditions for task initiation, termination, or branching, and there are several ways of dealing with this issue.

Effective planners depend on a proper domain representation. Different planners represent their domains differently, with different formalizations for capturing the information about the applications domain. Effective planners have built on the success of earlier planners. It is fair to say, however, that most of the planners reviewed by Tate et al. (1990) deal with deliberately abstracted and simplified domains in order to focus on the key planning issues and mechanisms. (See Wilensky [1981] for a general overview of how more realistic and complex domains might be dealt with through meta-planning.)

Planning is time constrained and resource dependent. Several planners have begun to attack the problems imposed by time and resource constraints in recognition of the need for "practical" solutions to real-world planning problems. Much of the work in efficient search algorithms has supported the implementation of "practical" planners. Although motivated by memory and learning considerations, case-based planners such as Chinese meal planning (CHEF)

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

(Hammond, 1986) may have significant potential for the eventual development of practical planners in knowledge-intensive domains.

Planning is situation dependent. Early planners were "open-loop" in that the world remained static once the plan was initiated. More recent planners deal with change during plan execution, as well as uncertainty in specifying the state of the environment. Interleaving of planning and execution is described by McDermott (1978) and replanning by Hayes (1975); general issues concerning situation-dependent planning are covered by Georgeoff and Lansky (1990) in their description of the procedural reasoning system (PRS).

Planners can learn. Most planners generate plans from scratch, but the better ones learn from experience. Clearly these are the planners most appropriate for a military human behavior representation. As noted above, case-based planners such as CHEF (Hammond, 1986) show potential for performance improvement with experience. The interested reader is directed to Tate et al. (1990) for a more in-depth discussion of these essential characteristics of artificial planners.

A slightly more recent review of artificial intelligence planners, again covering approximately 150 citations, is provided by Akyurek (1992), who proposes the basic planner taxonomy shown in Table 8.5. The key point here is the distinction between search-based and case-based planners: the former treat planning as a search in an appropriately structured state space characterizing the domain, while the latter treat planning as a retrieval and adaptation of prior plans and plan segments, using an appropriate feature space for case retrieval and an appropriate "adaptation" scheme for plan adjustment. Akyurek (1992) argues that the case-based approach appears to be better supported by the psychological literature and proposes the development and incorporation of a case-based planner within the Soar architecture (Laird et al., 1987). Fasciano (1996) builds upon

TABLE 8.5 Taxonomy of Planners (Akyurek, 1992)

Family

Type

Example

Reference

Search-based

Linear

STRIPS

Fikes and Nilsson (1971)

 

 

HACKER

Sussman (1975)

 

Abstraction

ABSTRIPS

Sacerdoti (1974)

 

 

NOAH

Sacerdoti (1975)

 

 

MOLGEN

Stefik (1981a, 1981b)

 

Opportunistic

OPM

Hayes-Roth and Hayes-Roth (1979)

Case-based

 

CHEF

Hammond (1986)

 

 

JULIA

Hinrichs (1988); Kolodner (1987)

 

 

PLEXUS

Alterman (1988)

 

SOURCE: Akyurek (1992).

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

the concept of case-based planning with his "Sim City" planner; here, the planner effects case adaptation by means of a qualitative causal model of the external world relations, which in turn is learned online through observation of past cause/effect correlations.

One other review of artificial intelligence planners should be noted, primarily because of its recency, if not coverage. A review by Kambhampati (1995:334) notes that "most implemented planners settle on some variant of the STRIPS action model," where the world is represented in a state space, actions transform the world from one state to another, and a plan is a series of such actions (Fikes and Nilsson, 1971). This approach runs into problems when the world is only partially observable, dynamic, and stochastic. Table 8.6 summarizes Kambhampati's (1995) observations on how current planners are attempting to deal with these problems.

TABLE 8.6 Advanced Artificial Intelligence Planner Techniques

World Characteristics

Planner Strategy

Fully observable, static, deterministic

Use classical planners

Partially observable

Specify information-gathering strategy

Dynamic

Interleave planning and acting

Stochastic

Use Markov models

A more recent although not as extensive review of cognitive architectures that includes planning functions is to be found at the following University of Michigan website: CapabilLists/Plan.html. Table 8.7 summarizes the architectures reviewed at this site, and the interested reader is referred there for further information.

TABLE 8.7 Example Planners as Part of Existing Cognitive Architectures

Planner

Reference

Gat's ATLANTIS

Gat (1991)

Theo

Mitchell et al. (1991)

Icarus

Langley et al. (1991)

Prodigy

Carbonell et al. (1991)

Meta reasoning architecture fax

Kuokka (1991)

Architecture for intelligent systems

Hayes-Roth (1991)

Homer

Vere and Bickmore (1990)

Soar

Newell (1990)

RALPH

Ogasawara and Russell (1994)

Entropy reduction engine

Drummond et al. (1991)

 

SOURCE: University of Michigan (1994).

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

From our review of these and other artificial intelligence planners, it would seem that many of the artificial intelligence planners are limited by a number of basic problems, including the following:

  • The environment with which the planner deals is often very simple (e.g., a blocks world), and the plan objectives and constraints are fairly trivial (e.g., stack the blocks in some order).

  • The environment is static, not dynamic, in the sense that it changes only in response to the planner's actions (e.g., route planning). Often, the planner is the only agent inhabiting the environment.

  • The environment is deterministic, not stochastic, so that the effects of the planner's actions are the same every time.

  • The planner has perfect information about the environment, with complete omniscience, error-free sensors, and perfectly timely feedback.

  • The planner works from first principles (e.g., means-end analysis), has little explicit domain knowledge, avoids useful domain-specific heuristics, and fails to learn from past successes and failures.

Clearly, the above limitations are most applicable to early planner designs; as noted earlier by Kambhampati (1995), more recent work has focused on ameliorating one or more of these problems. One last key shortcoming of artificial intelligence planners in general is due to the artificial intelligence planning community's interest in developing computationally effective planners, and not necessarily in developing models of human planners. Thus, most artificial intelligence specialists working on planner development efforts pay scant attention to how actual human planners solve problems. Modeling of human planning behavior per se is left to those in the behavioral science community who are interested in planning and problem solving. Some computational modeling efforts outside the military domain have been conducted, however, and the interested reader is referred to the behavioral science planning literature.

Our review of the literature has, unfortunately, uncovered remarkably few attempts at developing computational models of human planning behavior. Perhaps the earliest and best-known effort is the analysis and modeling of human errand-planning behavior performed by Hayes-Roth and Hayes-Roth (1979). They observed a mix of strategies used by humans, involving different levels of abstraction; partial planning; and, most clearly, opportunistic planning, a behavior that involves situation-based triggering of new goals, subgoals, and/or partial plans (e.g., "Well, as long as I'm in the neighborhood, I might as well go to the pet store nearby, even though it wasn't a high-priority item on my errand list."). The resulting opportunistic planning model (OPM) developed to reflect the observed planning behavior in the errand-planning domain makes use of a blackboard architecture (Hayes-Roth, 1985), which is reviewed by specialized (rule-based) agents to see if they can contribute to the plan, and if so, to which they post their plan components. The key architectural contribution of the OPM is the five conceptual levels of the blackboard (Hayes-Roth and Hayes-Roth, 1979):

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
  • Executive—determines goal priorities, attentional focus, and resolution across competing agents.

  • Meta-plan—determines the planning objective, problem representation, policies for constraint satisfaction, and plan evaluation criteria.

  • Plan abstraction—determines more specific plan intentions, schemes for meeting these intentions, overall strategies, and specific tactics to be used.

  • Plan—determines specific plan goals, a plan design to meet these goals, general procedures for action, and detailed operations for finer action.

  • Knowledge base—maintains key problem-specific planning information, such as subgoal priorities and route geometry.

An implementation of this architecture, populated with about 40 rule-based agents, allowed Hayes-Roth and Hayes-Roth (1979) to compare OPM-generated protocol and plans with those generated by human subjects. As noted by the authors, some sections of the protocol (i.e., the behavior engaged in while generating the plan) showed remarkable similarities, while others diverged; the resulting plans, however, were not too far apart. Overall, the OPM seems to be a reasonable candidate for modeling human behavior, at least in this domain. Further validation work would appear necessary, however.

A partial validation of this approach is provided by Kuipers et al. (1988) in their study of how pulmonary physicians develop diagnosis and treatment plans (i.e., information-gathering and action plans) for treating a patient in a high-risk, uncertain medical scenario. Kuipers et al. conducted a protocol analysis during treatment plan generation ("thinking aloud") and after specification of a plan ("cross-examination") with a number of medical specialists to develop a behavioral database. They then attempted to match the protocol and treatment plan using a classical decision-theoretic model approach, which provides a rational means for combining diagnostic uncertainty with treatment utility (see Chapter 6 for further discussion). They found, however, that this model fails to capture the planning behaviors they observed. Instead, Kuipers et al. (1988:193) found:

The decision process we observed in the protocols is well described as one of planning by successive refinement of an abstract plan (Sacerdoti, 1977), combined with opportunistic insertion of plan steps (Hayes-Roth and Hayes-Roth, 1979).

Kuipers et al. go on to suggest (but not implement) a simple rule-based model to explain the protocols they observed and to argue, more generically, that such a planning model:

  • Is simpler to effect by a human than one requiring an estimate of probabilities, a specification of utilities, and a computation of expected value.

  • Is simpler to remember, at least at high abstraction levels (e.g., the Hippocratic dictum "First, do no harm.").

  • Can be explained and defended (to patients and colleagues) most easily through a step-by-step syllogistic explanation.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
  • Can be built up from experience through rule abstraction from past cases.

We are unaware of related follow-up work focusing on the development of a computational planning model. It would appear, however, that the behavioral data and initial moeling efforts of Kuipers et al. (1988) would support such work.

CONCLUSIONS AND GOALS

Short-Term Goals

  • Make explicit the difference between doctrinal and actual planning behaviors. Modeling the former behaviors may be easier since they are well documented, but the results may not be representative of actual behavior. Modeling the latter may yield a more accurate representation of what occurs, but several years of research will be required to measure, describe, formalize, and formulate such models.

  • Begin the above research process to build the behavioral database needed to support follow-on development of nondoctrinal planning models.

  • In parallel, for the short term, develop planning models based on doctrine. Some of the observed (nondoctrinal) behaviors may be reproduced by these models because of limitations in "upstream" process models (e.g., perception, situation assessment); other behaviors might be modeled by injecting deliberate deviations from doctrine through, say, a fuzzy logic implementation of a "crisp" doctrinal rule.

  • Consider expanding the scope of most military planners to encompass the entire planning process, from receipt of command orders, to issuance of subordinate orders, to plan monitoring and replanning.

  • Give greater consideration to the next generation of planning decision aids that will be deployed on the battlefield. Some of the lower-level planning functions (e.g., route optimization) may not need to be modeled within the human behavior representation because of their automation (analogous to the drop-off in pilot modeling research as autopilots were introduced).

  • Consider expending effort on developing models in the tactical domain from the ground up, rather than searching in other domains for candidate models to adapt to the military case. It is not clear that any particular community has the answer to the development of sophisticated planning models, algorithms, or decision aids, especially at the higher conceptual levels of plan generation (although there are many domain-specific tools to support the more tedious plan elaboration tasks).

  • Consider expending development effort on codifying the large numbers of doctrinal (or nondoctrinal) planning behaviors once those behaviors have been agreed upon. The conceptual model of the mission space (CMMS) development effort may be an appropriate place to start. It does not appear likely that a "first-

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×

principles" planning architecture emphasizing process sophistication over knowledge content will succeed in the highly knowledge-intensive and options-constrained military environment.

Intermediate- and Long-Term Goals

  • Devote effort to developing planners that incorporate, at the least, an extensive domain-specific knowledge base and, more important, a learning capability to build upon past planning and operations experience.

  • Develop planning models that are more reactive to account for plan failures, dynamically changing environments, and changes in plan goals. The focus should be on developing planners that are more robust to failures, develop contingency plans, and can rapidly replan on the fly.

  • Develop planning models that account for a range of individual differences, skill levels, and stressor effects. Since the military plan is such a key determinant of overall mission success, it is critical that this behavior be represented accurately across the population of military planners.

Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 203
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 204
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 205
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 206
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 207
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 208
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 209
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 210
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 211
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 212
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 213
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 214
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 215
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 216
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 217
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 218
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 219
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 220
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 221
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 222
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 223
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 224
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 225
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 226
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 227
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 228
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 229
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 230
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 231
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 232
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 233
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 234
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 235
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 236
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 237
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 238
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 239
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 240
Suggested Citation:"8 Planning." National Research Council. 1998. Modeling Human and Organizational Behavior: Application to Military Simulations. Washington, DC: The National Academies Press. doi: 10.17226/6173.
×
Page 241
Next: 9 Behavior Moderators »
Modeling Human and Organizational Behavior: Application to Military Simulations Get This Book
×
Buy Paperback | $110.00 Buy Ebook | $89.99
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

Simulations are widely used in the military for training personnel, analyzing proposed equipment, and rehearsing missions, and these simulations need realistic models of human behavior. This book draws together a wide variety of theoretical and applied research in human behavior modeling that can be considered for use in those simulations. It covers behavior at the individual, unit, and command level. At the individual soldier level, the topics covered include attention, learning, memory, decisionmaking, perception, situation awareness, and planning. At the unit level, the focus is on command and control. The book provides short-, medium-, and long-term goals for research and development of more realistic models of human behavior.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    Switch between the Original Pages, where you can read the report as it appeared in print, and Text Pages for the web version, where you can highlight and search the text.

    « Back Next »
  6. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  7. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  8. ×

    View our suggested citation for this chapter.

    « Back Next »
  9. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!