Dempster-Shafer Evidential Theory Belief Amalgamation and Dynamic Programming Supporting Soldier Squadron Adversarial Engagement: Simulation-Based Decision-Making

Increasing complexity of adversarial engagement in urban warfare scenarios suggests that there is a serious demand for technologies that can assist the warfighter's decision-making capabilities.This is especially true for stressful conditions experienced by soldiers where optimal and expedient execution of missions and directives is needed under the yoke of resource constraint.In such situations, the age-old military and historical hallmarks of squadron group belief consensus supporting mission accomplishment and optimal movement of a unit based on probabilistic assessment of value come to the fore and are vital elements in supporting and preserving mission integrity.Both fundamental drivers for squadron success can be supported and demonstrated using a two-element statistical machine learning formulism.The first part utilizes Dempster-Shafer evidential theory for belief amalgamation across a squadron and the second part dynamic programming for estimation of optimal path movement through a hostile domain.Simulations using these two important geo-intelligence processors are performed to demonstrate machine learning-based processing which can ultimately be utilized in intelligent squad weapon systems aiding in decision-making.The first formulism is developed around the geo-intelligence scenario of a 4-soldier squadron trying to decide which direction to proceed based on the amalgamation of individual beliefs.Dempster-Shafer evidential theory provides a way to amalgamate the beliefs of each soldier via a method that uses orthogonal summation of probability mass associated with different propositions.Eleven propositions exist for the 4-directional problem encompassing not only the initial 4 directions, but propositions based on logical 'or' as well as the state of ignorance parameterizing uncertainty.An orthogonal amalgamation template allows for amalgamation of belief information for soldiers 1 and 2 and then soldiers 3 and 4.These soldier beliefs are then amalgamated into a virtual soldier for all soldiers representing the squadron group mind.Simulation results demonstrate that given the initial probability mass profiles for each soldier, direction 2 is the optimal direction to proceed in providing a logical guide for squadron movement.The increasing complexity of warfighter scenarios with respect to adversarial engagement and non-traditional environments of contention also suggests a serious need to leverage technology to go beyond mere state estimation towards algorithms prescribing actions.Dynamic programming is well suited for policy-based decision-making where there is a need to assist humans in making decisions where the best choices or actions are not clear and depend on values placed on specific situations or states.To demonstrate the applicability and power of dynamic programming to path-based decision-making, a fictious model problem of finding the optimal policy for moving a soldier squadron through a multiple room building is addressed.The objective is to illustrate how prior geo-intelligence information in the form of transition probabilities and rewards can be used to facilitate decision-making in terms of what should optimally be done rather than what must be done.The objective is also focused on understanding the effect of noise infiltration on optimal policy estimation. 103-2A soldier squadron is tasked with moving through a 4-floor building and 'clearing' it.A 16 state-2 action dynamic programming model for building 'clearing' is created where the aim is to understand on average what rooms should be places of hostile engagement (action 1) and what rooms should not (action 2).Prior ground clandestine geo-intelligence provides the room-to-room transition probability and reward fields.The geo-intelligence dynamic programming issue of interest is how the optimal policy (function mapping of state to action) changes as noise infiltrates the geo-intelligence system, changing the transition probability and reward fields.Baseline transition probability and reward fields for the 16 state-2 action domain are set up where particular attention is paid to their prescription based on underlying fire fight capability.The optimal policy resulting from the baseline transition probability and reward fields demonstrates that hostile engagement is optimal along a floor while passivity is optimal at the floor transition points.This is consistent with floor transitions being choke points where quick movement rather than fighting is more highly rewarded.When the reward field for hostile engagement is lowered near the end room floor transition points, continual hostile engagement completely along a floor is no longer optimal.Hostile engagements should be performed near the beginning and middle of the floor.Noise infiltrating prior clandestine geo-intelligence causes modulation of the state transition probability field which cause changes in the baseline optimal policy.Results suggests that uncertainty in the transition probability field associated with one type of action precipitates increases in the alternate action.This suggests that increases in hostile engagement actions are associated with increased uncertainty in transition probabilistic information for the non-hostile engagement action.Increases in non-hostile engagement action are associated with increased uncertainty in transition probabilistic information for the hostile-engagement action.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.