Hill, Surdu & Pooch - Anticipatory Planning Support System
PAEI_033_Hill_Surdu_Pooch_APSS.gif

The Structure of Concern Project compares many theoretical models from many disciplines to the Adizes PAEI model, arguing that they must all be reflecting the same underlying phenomenon. One concern structure model is described below.


There is a military saying that a plan never survives the first shot. Military planners know and expect that unanticipated events will overtake their plans, forcing commanders into a reactive planning mode during operations. Hill, Surdu and Pooch propose a system architecture called the Anticipatory Planning Support System (APSS) to support planners as they update their plans in real time by helping them anticipate enemy and friendly courses of action as news comes in (Hill et al., 2000; Hill & Surdu, 2001).

APSS is an agent-based decision support system that works with another such system called OpSim (Operationally-focused Simulation, Surdu & Pooch, 2000). It operates in a similar fashion to APSS, but it tracks operations as they deviate from plan, to alert commanders about possible ramifications of those deviations. The two systems exercise similar functions in different domains: one anticipates the impact of contingencies and the other anticipates the consequences of operational drift.

The APSS Execution Monitors digest live information to revise plans and expectations on the fly, which is a P function of adjusting as you go along. OpSim Operations Monitors process live information to monitor deviation from plans, which is an A function of scanning for errors. There is a Plan Generator that explores the revised problem spaces for new opportunities and threats, which is an E function of projecting hypothetical possibilities. The I function is headed by a Planning Executive agent which coordinates the activities of the above three agent types as well as models of the outside world (the World View) and communications with the outside world (World Integrator). The Planning Executive is also the level at which human users interact with the system. Thus all elements of the system are integrated into one ensemble by the Planning Executive. The four agent types in this system are described more fully below.

P – Execution Monitors: (APSS) Anticipates the immediate future by attaching to a node in the plan and assessing the difference between actual and planned states for that node. Derives anticipated states by forward simulation along appropriate branches, determining the likelihood of each anticipated state. Execution Monitors essentially explore an option space at a faster rate than the outside world does.

A – Operations Monitors: (OpSim) A dynamic hierarchy of rational agents, each of which monitors a specific aspect of the plan, comparing the actual operation with the planned operation. Significant deviations trigger impact assessment processes, and the top-level OM informs the decision maker when the plan is in jeopardy. Operations Monitors thus generate an error signal or learning signal for a negative feedback process, comparing the actual state of the world to a reference state (the plan).

E – Planner: (APSS) This agent reads the state of a node and uses a Branch Generator to consider possible future actions and produce new relevant branches. The Branch Generator uses simulations, inference mechanisms and genetic algorithms guided by the goal-states or end-states desired by the various parties to the conflict to generate possibilities. A Branch Evaluator then determines how well the terminal node of the new branch accomplishes friendly goals, using simulations and inference. This process generates the option space and the reward or utility (cost/benefit value) of each option.

I – Planning Executive: This agent embodies all four concern specializations, but its role is integrative. It coordinates the flow of tasks and information, allocating certain plan nodes to planners or monitors as required, prioritizing tasks given system resource constraints, and basically cooperating with and controlling all areas of the system as initialized by World View information to make everything work together as an organic and evolving whole.

Bibliography
1. Hill, J. M. D., & Surdu, J. R. (2001). “Simulation And Agent Cooperation In Dynamic Plan Building.” Pennsylvania State University & NEC Labs.
2. Hill, J. M. D., Surdu, J. R., & Pooch, U. W. (2000). “Anticipatory Planning Using Execution Monitoring And A Constrained Planning Frontier.” Pennsylvania State University & NEC Labs.
3. Surdu, J. R., & Pooch, U. W. (2000). “Simulation Technologies in the Mission Operational Environment.” Simulation, 74(3), 138-160.
Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License