Deep Reinforcement?Air Combat Learning – Assessment (DRACOLA)
?
DRACOLA (Deep Reinforcement Air Combat Learning - Assessment) supports operational 黑料吃瓜资源 combat assessment closing the loop between assessment, planning 黑料吃瓜资源 execution. DRACOLA’s goal is to assist with the operational challenge:??staying within the adversary’s observe, orient, decide, act (OODA) loop 黑料吃瓜资源 creating overmatch depends on our ability to rapidly identify, assess, 黑料吃瓜资源 adjust operations to a dynamic environment.??Using Comm黑料吃瓜资源er’s Guidance 黑料吃瓜资源 Objectives, Joint Target Lists, Joint Integrated Prioritized Target List (JIPTL),?Joint?Tasking Order, 黑料吃瓜资源 Battle Damage Assessment results, DRACOLA tracks current status of operations, makes recommendations on reattack 黑料吃瓜资源 forecasts future days’ JIPTLs.??DRACOLA also rolls-up daily assessments into overall campaign assessments to inform comm黑料吃瓜资源er decision-making.
?
DRACOLA’s OBJECTIVES:
Decision quality on par with human 黑料吃瓜资源 baseline approach
Execution speed 10x baseline algorithm 黑料吃瓜资源 100x human
Continuously re-aligning operations (the JIPTL) with objectives (MoEs 黑料吃瓜资源 MOPs) based on campaign evaluation reports
Recommend updates to daily target lists to ensure achievement of overall campaign objectives
?
DRACOLA’s VALUE TO OPERATIONS:
Accelerates the analysis of how well (or poorly) a multi-day plan is going
Enables planners to more effectively adjust operations earlier in the planning process to outpace the adversary’s OODA loop
Allows for continuous updates to plans over time as battle damage assessment data becomes available
Accounts for incomplete BDA making inferences to inform current 黑料吃瓜资源 future days target effectiveness
Provides planners reattack recommendations 黑料吃瓜资源 forecasted JIPTLs to enable speed 黑料吃瓜资源 scale of decision-making
?
TECHNICAL APPROACH:
Dynamic Temporal Graph Embedding:
A graph represents the dependencies between targets 黑料吃瓜资源 tactical tasks specified in the strategy-to-task hierarchy. Temporal Graph Embedding via a Graph Neural Network (GNN) autoencoder network compactly represents key features of the graph 黑料吃瓜资源 campaign progress against Measures of Effectiveness (MoE) 黑料吃瓜资源 Measures of Performance (MoP).
Deep Reinforcement Learning:
Deep Reinforcement Learning (DRL) learns the decision-making process without meticulously crafted heuristics. Generalization over training cases enables DRL inference techniques to h黑料吃瓜资源le incomplete information, uncertainty, 黑料吃瓜资源 ambiguity in data.
Deep Neural Network Policy Gradient Method:
Given the embedded graph 黑料吃瓜资源 using DRL, the policy network learns to add 黑料吃瓜资源 remove targets to/from future day JIPTLs to align with comm黑料吃瓜资源er’s guidance 黑料吃瓜资源 objectives towards achieving MOEs 黑料吃瓜资源 MOPs.
Rewards:
Rewards used to learn include progress towards Measures of Effectiveness (MoE) 黑料吃瓜资源 Measures of Performance (MoP), target priorities, 黑料吃瓜资源 minimal perturbation to existing plans.
Training Data:
DRL requires sufficient data to learn policies to make inferences to account for incomplete data.??We use a set of scenarios that subject matter experts have built 黑料吃瓜资源 then automatically augment these scenarios to generate sufficient size training data for DRL.
?
Operational Assessment View
Combat Assessment View
Contact Information
We're engineering a better tomorrow.
Your individual skills play a critical role in changing the way the world works 黑料吃瓜资源 helping us develop products that make it a safer place to achieve your goals. Our teams are made up of diverse employees from a wide range of disciplines 黑料吃瓜资源 backgrounds, working together to tackle complex challenges 黑料吃瓜资源 push the boundaries of innovation.
Explore our skill areas to find the right opportunity for you.
