Anticipatory Planning Support System Dr John R Buck

Anticipatory Planning Support System Dr. John R. “Buck” Surdu Major, U. S. Army Senior Researcher, ITOC West Point, New York John M. D. Hill Lieutenant Colonel, U. S. Army Assistant Professor, USMA Ph. D. Candidate, TAMU Dr. Udo W. Pooch Department of Computer Science Texas A&M University College Station, Texas

Agenda • Introduction • How we plan now and what we can do about it • The Anticipatory Planning methodology • A simple scenario • Some ideas from the scenario • A simple example of the methodology • Status of the research • Conclusion

How we plan now • Traditional Military Decision Making Process (MDMP) focuses on developing a few friendly Courses of Action (COAs) against the “most-likely / mostdangerous” enemy COAs • This results in a very detailed plan that considers only a few branches • There is a well-known axiom in the military that “No plan survives the first shot” - this is another way of saying that a branch has occurred during execution which was not included in the plan • When this happens, the commander and the staff immediately transition into reactive mode

What can we do? • The Army needs a new way to perform planning while executing the operation so that it retains “option dominance” • General (ret. ) Wass de Czege has proposed an entirely new approach to military planning and execution called “Anticipatory Planning” that merges planning and execution • As many branches as are reasonably possible are developed in the initial process, and as execution progresses the plan is continuously updated based on actual events • Future branches that are known to be invalid are nominated for pruning and new branches are developed – well before they occur in execution • Planning effort is expended on the most likely or most valuable branches • Computational power, modern planning techniques, and rapid simulation can be applied to this problem to produce a powerful new planning and execution methodology and support system

Proposed Methodology Planning Executive Data from the current, real operation (e. g. , ABCS, GCCS, etc. ) Execution Monitors World Integrator Plan Description Branch Evaluator Planners Branches Generator Real information plus “dead reckoned” information World. View Plan Information Control Information

Major Components • World View – Keeps track of the Actual State of the operation • Plan Description – Represents the possible ways the operation can proceed – Nodes retain the Planned States; Branches have pre-conditions • Planning Executive – Monitors use of system resources and determines the priority of planning – Controls the instantiation and activities of Execution Monitors and Planners – Constrains the Planning Space and the Planning Frontier • Execution Monitor – Conducts forward simulation from the Actual State to produce an Anticipated State at its monitored Node – Determines the significance of differences between the Anticipated State and the Planned State at that Node, then makes a recommendation to the Planning Executive if re-planning is appropriate • Planner – Invokes a Branches Generator to develop significant, representative Branches – Invokes a Branch Evaluator to determine Viability and Likelihood Measures

Tactical Entities • Entities have attributes and can be assigned tasks – Units, Indirect Fires, Obstacles, etc. • A Plan Description is composed of many actions taken by the entities • Several different types of tasks – Place, Idle, Move, Engage, Etc. • Some Tasks are synchronized with other Tactical Entities – Engage, Coordination, Rendezvous • Some paths (chain of actions) through the Plan Description are better than others (in terms of how well they accomplish the mission)

Node • The node has an “earliest time” measure based on the sequence of Branches from the Initial State • The Node contains a mapping between Tactical Entities and their state at that Node in the Plan Description • Each state contains the current status of relevant attributes – Location, strength, fuel remaining, ammo remaining, etc.

Branch • Creates and maintains a Task List for each Tactical Entity that appears in the Planned State of the previous Node • Each Task List maintains an ordered list of the Tasks the Tactical Entities must perform • The possibility of an entity choosing a different task results in branching from that Node • Branches represent the flow of the battle for each choice

Simple Methodology Example Actual State of the Operation Node with a Planned State Node with an Execution Monitor

Invalidation of Branches Invalidated Branch Node with a Planner Actual State of the Operation

Re-Planning Newly Generated Still Invalidated Newly Invalidated Re-Validated Actual State of the Operation Nominated for pruning

Planning Space and Planning Frontier Planning Space Planning Frontier Actual State of the Operation

Simple Scenario • Blue Force mission: – Secure the Eastern Pass – Don’t let the Red Force secure the Western Pass • Red Force mission: secure the Western Pass • There is a time constraint

Plan One • Red Force actions – Wait at the Eastern Pass • Blue Force actions: – Wait at the Western Pass • Plan Quality (Blue perspective) – – Red does not secure Western Pass No Casualties FAILS to secure Eastern Pass Unrealistic Red option Plan Description Task Lists

Plan Two • Red Force actions – Move south to the Western Pass • Blue Force actions: – Move south, engage Red, move to the Eastern Pass • Plan Quality (Blue perspective) – – Red does not secure Western Pass Acceptable(? ) casualties Secures the Eastern Pass More realistic Red option Plan Description Task Lists

Plan Three • Red Force actions • Plan Quality (Blue perspective) – Move north to the Western Pass • Blue Force actions: – Move south, engage Red (oops!) – Move to the Eastern Pass – – Red secures Western Pass No casualties Secures the Eastern Pass More intelligent Red option Plan Description Task Lists

Simulation in Plan-Building • As the human planner builds a plan • Cannot produce or store every possible state • Can describe a plan from start to finish as a sequence of tasks (some of which are independent, others of which are interdependent) • Can establish critical states at which different options (sequences of tasks) can be taken • Several rules are enforced by system • Simulation used to – Determine Resource expenditures (time, fuel, ammo, casualties) – Provide feedback regarding viability – Determine the Planned State at the end of a Branch

Simulation in Execution Monitoring • Execution Monitor (EM) is placed on a node • The EM examines the planned state at that node • The EM subscribes with the Plan Description (PD) for updates about the status of every Tactical Entity (TE) involved in the planned state • The PD maintains a list of interested EMs for each TE, and publishes updates to all of those EMs whenever a status change occurs • Each EM can restrict the simulation to executing the tasks of a specific TE (and all TEs that are involved in Synchronized Tasks with that TE)

Simulation in Re-Planning • After the Planning Executive accepts a recommendation for re-planning from an Execution Monitor • The Branches Generator uses several algorithms and heuristics to produce alternative Branches • The Branch Evaluator uses simulation to determine the outcome of those alternative Branches • The most-likely (whether favorable or unfavorable) and most-representative Branches are retained • If you can’t plan all the way to completion, a local measure of the viability of a Branch is a useful heuristic • There are very many possible branches, but in reality there are usually only a few good ones

Simulation to Drive the System • Test Driver uses a Plan Description and a simulation to provide inputs into the system about the “Actual State” • Each Execution Monitors in the APSS uses simulations to “move” the Actual State to its monitored Node, creating Anticipated States • When APSS changes the Plan Description, changes must be propagated to the Test Driver’s Plan Description Actual State Test Driver APSS Plan Description

Plan Builder Prototype

Conclusion • The Anticipatory Planning process accounts for the chaotic nature of warfare in which possibilities appear and disappear • The purpose of this research is: – To maintain as many viable options in the plan as possible – To focus planning effort where it will do the most good – To help the commander and staff identify decisions that they { can | should | must } make NOW to ensure a desired option is available LATER • The Anticipatory Planning process, aided by an automated support system built on the Anticipatory Planning methodology, will help U. S. forces achieve option dominance, providing a decisive advantage

Questions?

Planning Techniques • Contingency Planning – Develop and consider every possible branch – Not feasible for large planning spaces • Probabilistic – Assign probabilities to branches and consider the most likely – Still have to develop a lot of branches – Sensitive to changes in the probabilities • Interleaved Execution and Planning – Observe the results of execution, then plan based on the new information – Not a good idea to wait to find out what happens to start planning • Reactive Planning – Develop reaction rules for specific conditions – Difficult to build enough rules to cover all possible conditions