Ontological Support for Living Plan Specification, Execution and Evaluation

Maintaining systems of military plans is critical for military effectiveness, but is also challenging. Plans will become obsolete as the world diverges from the assumptions on which they rest. If too many ad hoc changes are made to intermeshed plans, the ensemble may no longer lead to well-synchronized and coordinated operations, resulting in the system of plans becoming itself incoherent. We describe in what follows an Adaptive Planning process that we are developing on behalf of the Air Force Research Laboratory (Rome) with the goal of addressing problems of these sorts through cyclical collaborative plan review and maintenance. The interactions of world state, blue force status and associated plans are too complex for manual adaptive processes, and computer-aided plan review and maintenance is thus indispensable. We argue that appropriate semantic technology can 1) provide richer representation of plan-related data and semantics, 2) allow for flexible, non-disruptive, agile, scalable, and coordinated changes in plans, and 3) support more intelligent analytical querying of plan-related data. Keywords—adaptive planning; outcomes assessment; ontology I. THE NEED FOR ADAPTIVE PLANNING “No plan survives first contact with the enemy” (Clausewitz, On War). Real world uncertainties all but guarantee that even the most carefully developed plan will not be carried out exactly as intended. The military response, as in the business domain, has been to increase the speed and agility of planning and execution [1-4]. On the strategic level, the transition from the Joint Operation Planning and Execution System (JOPES) to an Adaptive Planning and Execution (APEX) system exemplifies this trend. In addition to speeding up the deliberate planning and review cycle, these efforts seek to increase the number of planned options and contingencies. According to the Adaptive Roadmap II, signed by the Secretary of Defense in March 2008, the ultimate goal is to provide plans that are “maintained continuously within a collaborative environment” to reflect any changes that impact any significant aspects of a plan. Such plans will together form something the Adaptive Roadmap calls a “living plan.” Plans may need to be adjusted to maintain their relevance based on changes in the world (e.g., weather, location of enemy troops, troop readiness, air assets). Additionally, they may need to be adjusted in order to maintain their coherence within a system of plans, such as when the goals of supporting or supported plans change. II. THE IDEA OF THE LIVING PLAN In the current state of military planning – as encapsulated in Joint Doctrine (JP 5.0) – a distinction is drawn between deliberate planning and crisis action planning. Deliberate planning is supply driven. Plans are static information objects created as the outputs of a deliberative, rule-governed process, and stored in a repository until needed. They may be created years ahead of actual use, or they may never be used at all. Crisis actions plans are demand driven: something happened and we need an urgent response; because the response should involve a degree of organized action, planning is needed. Crisis action planning is a response to the uncertainty involved in our knowledge of real-world states. But even deliberate planning rests on an institutional acknowledgement of our inability to accurately predict the future, in that Doctrine allows the making of ad hoc resource requests which deviate from the deliberate plan as specified. Sometimes, on first contact with the enemy, deliberate plans break and workarounds are needed. Regardless of the quality of the prior deliberation that went into the deliberate plan, the need for such corrective actions as a result of the unanticipated interactions between blue forces and the world make for suboptimal procedures. The goal of the living plan is to remove this ‘breaks because it would not bend’ feature of the deliberative plan by minimizing the distinction between deliberate planning and crisis action planning through a new type of planning process that is marked by constant update in light of updates in our real-world knowledge. The idea is to embed into the very fabric of plan representation our uncertainties about the world, so that the activity of planning is transformed from one of the creation of plans as outputs to a process of continuous plan development. The living plan itself becomes a probabilistic, branching information artifact – a representation of the moment-to-moment intentions not merely of single platoon commanders but of the military as a whole. It incorporates at each phase representations of multiple alternative courses of action which are continuously changing in light of actual and projected states of the world, adjacent plans, supporting and supported plans. III. ADAPTIVE PLANNING REQUIREMENTS We believe that any computational approach to supporting the Secretary of Defense’s goal for living plans must meet six critical requirements. First, it must be able to represent all the types of entities and relationships, knowledge about which is important to maintaining a living plan. This requires a highly expressive representational capability to capture, manage, and reason over plans, plan elements (e.g., goals, available assets, weather, battle terrain), and their relations within a system of plans. Second, any approach must be able to detect meaningful changes that impact plan relevance and coherence. This requires effective monitoring and sensitivity analysis to identify in a reliable and scalable way those changes which are of significance to the system of plans [5,6]. Recognition of the significant changes must then trigger processes that maintain the relevance and coherence of this system at multiple levels and across plan elements. Third, any approach requires coordinated adjustment processes, which are needed to fulfill the second requirement (above). Such processes must be able to run independently, be applicable (when necessary) to real-time conditions, and be capable of harmonizing with other large-scale plan adjustments. Fourth, any such approach requires automated information extraction and routing because maintaining realistic plans requires more information processing than can be achieved through manual methods alone. Fifth, whether in support of human planners, warfighters during mission execution, operations assessment staff, or automated systems performing the same tasks, any approach needs to support analytical queries against the ensemble of plan-related data. Since plan-related data is very heterogeneous, this amounts to applying a unified structured query front end to structured and unstructured data on the backend. Sixth, joint warfighters at all levels of command will need to collaboratively plan and execute in conjunction with semiautomated adaptive planning systems. Therefore, any approach for providing living plans must support extensible and versatile interactive applications that can deal with the sorts of diverse but integrated user environments required for living plans. Relative to the six requirements described above for supporting the Secretary of Defense’s goal for living plans, our overall approach is based on the idea that semantic representation of data by means of ontologies, combined with probabilistic classifiers operating in a transactional environment, will allow the needed representation, monitoring, analysis, sharing and querying of information at distinct levels of granularity and detail and across distinct applications. The system will be required, for example, to display a JFACC’s view of ATO mission plans, a squadron Commander’s view of the day’s mission plan, and STRATCOM’s view of a Theater. As in other domains, the semantic approach is designed to reduce information siloes, and enable effective tailoring of knowledge and information to different needs. It is designed also in such a way as to allow incremental improvements over time, as shortcomings in the framework uncovered at any given stage are rectified in subsequent stages. In what follows we focus on the first and fifth requirements described above: for rich representations of data and semantics, and for the capacity to use such representations in mounting queries against plan-related data. As regards the former, we describe the coverage domain of our proposed Plan Ontology (see Figure 2) in terms of how we: (a) model plans in terms of cyclical phase-specific attributes; (b) embed metrics that relate plans to world conditions; and (c) embed meta-metrics that use the metrics under (b) to create an incremental plan and plan-execution improvement process across the whole system. On each level multiple families of related terms will be required, including definitions and axioms specifying the relations between them. As regards the latter, we describe how queries are passed through parts of the system in order to illustrate some of the semantic relations that need to be computed in order to support analytically useful queries over living plan data. Figure 1: Draft of the Plan Ontology at http://ncor.buffalo.edu/plan-ontology IV. REPRESENTING PLANS IN RECURRING PHASES To better understand and support the notion of continuous, living plans, we require a view of planning that is more abstract than is traditionally employed. The simplistic notion of ‘the plan’ created prior to ‘the execution’ is at odds with our view of planning as a dynamic, continuous, iterative process that not only adapts to the effects of planned actions, but also adapts the process of planning itself in ways designed to achieve more satisfactory outcomes over time. Our model focuses on three primary factors in the planning process: 1. different phases of the planning process (successive phases within a given course of planning processes), 2. types of judgments within each of those phases that enable effective planning, and 3. information, including metrics, on which these judgments are grounded. On the traditional view, planning only happens periodically as a precursor to its execution. Here, in contrast, we view the total planning process computationally as forming a series of parallel, interacting courses or flows at a number of different levels. These processes unfold dynamically, with changes in any given course being communicated to parallel and hierarchically related courses wherever changes in the latter are required. The system is organized in such a way that updated versions of needed plans and subplans can be generated at any point in time. Each parallel course is itself seen as being organized into a succession of three phases corresponding roughly to the first three phases of the well-known Plan Do Check Act (PDCA) cycle, and similar models. A difference is that the phases in our framework are viewed as continuous and intermeshed with each other rather than discreet. Especially the Act phase, where adaptive actions are taken, is distributed and continuous across the other phases.  development – This phase consists of processes of identifying, considering, selecting, constructing, and/or modifying potential courses-of-action (COAs) that are expected to satisfy a goal. This includes the process of creating and maintaining potentially executable ‘plans sitting on the shelf’ in traditional, deliberate planning – referred to in our ontology as ‘plan specifications’. The distinguishing feature of this phase is that there has been no decision to take actual actions in conformity with and under commitment to any specific plan.  execution – This phase involves processes of planning while acting according to a particular planned COA. Unlike random or spontaneous actions, such planned processes can be evaluated relative to the plan. For example, indicators can be used to judge whether the intermediate effects of planned actions are consistent with expectations. But, as the plan has not yet terminated, the net effect of all planned actions relative to the goal set forth in the plan cannot be judged. A key planning process in the execution phase is the making of a decision to terminate execution because the goal has been achieved, or because the plan is no longer relevant or coherent, or is being executed unsatisfactorily.  post-execution – This phase involves the postexecution processes of interpreting and judging an executed plan and its outcomes relative to expectations. In this process, all actions taken under commitment to the plan have been taken. Thus their net effect can be assessed relative to the specified goal. The primary purpose of the processes involved in this post-execution phase is to enhance future planning, for example by: o defining new goals; o clarifying existing goals; o improving effectiveness in achieving goals. Associated with processes of each of the mentioned types are four basic planning-related judgments that enable reasoning aimed at leading to the creation and selection of better plans:  relevance – How well does the current state of planning relate to actual or anticipated external world conditions, such as constraints, opportunities, planned outcomes, unplanned side-effects, etc.?  coherence – How well do the processes of planning on-going in the current phases relate to other synergistic planning processes. In other words, are they in conflict or coherent with other friendly force, coalition, political, etc. planning?  planning-assessment – How well were the processes in each phase of planning performed by the planner, from a single person to an organization?  meta-metric learning – How well does the current set of metrics support the goal of evolutionary improvement of the entire planning process (and, as a consequence thereof, the entire process of creating and executing and evaluating plans)? V. REPRESENTING RECURRING CLASSES OF METRICS IN SUPPORT OF CYCLICAL PLAN PHASES In this section, we bring together the three factors of planning outlined above – phases, judgments, and metrics – to see how they merge to form a more complete picture of a continuous adaptive planning process. For each combination of planning phase and judgment we provide example metrics. These are provided here for illustrative purposes only, and especially as concerns plan execution our framework will draw on the extensive list of Measures of Effectiveness and Performance identified in salient doctrine for the tasks of the Universal Joint Task List, for example as described at: http://www.dtic.mil/dtic/tr/fulltext/u2/a398683.pdf A. Plan Development Phase