Multi-Aircraft Behavior Management With ML and Rule Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing aircraft simulations face challenges in managing large groups of aircraft with high realism and accuracy, requiring significant computing resources and struggling with complex interactions and behaviors.
Innovation Solution
A hierarchical system combining a machine learning model and a rule system to control vehicle behavior, where the machine learning model selects behaviors and the rule system executes actions, with an action mask to prevent invalid choices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional multi-agent based simulations are used to model large groups of aircraft, then the simulation can capture individual aircraft behaviors and interactions, but the computing resources required become excessive and the system becomes unmanageable
Solution Approach 1:
The patent segments the control system into two distinct components: a machine learning model that determines high-level maneuver behaviors and a rule system that executes specific actions. This segmentation allows the complex simulation to be managed by distributing computational tasks - the ML model handles strategic decisions while the rule system handles tactical execution, reducing overall computing resource requirements while maintaining simulation accuracy.
Solution Approach 2:
The patent introduces an action mask as an intermediary component that filters and validates potential actions before execution. The action mask receives candidate actions from the machine learning model and selectively allows or blocks them based on predefined criteria. This intermediary layer reduces computational overhead by eliminating invalid actions early in the process and ensures only feasible maneuvers are executed, thereby managing computing resources more efficiently.
2Productivity
If a machine learning model is used to select maneuver behaviors, then the speed and accuracy of behavior selection is improved, but the system complexity increases
Solution Approach 1:
The patent divides the control system into two distinct components: a machine learning model that determines high-level maneuver behaviors and a rule system that executes specific actions. This segmentation allows the complex simulation to be managed by distributing computational tasks - the ML model handles strategic decisions while the rule system handles tactical execution, reducing overall computing resource requirements while maintaining simulation accuracy.
Solution Approach 2:
The patent introduces an action mask as an intermediary component that filters and validates potential actions before execution. The action mask receives candidate actions from the machine learning model and selectively allows or blocks them based on predefined criteria. This intermediary layer reduces computational overhead by eliminating invalid actions early in the process and ensures only feasible maneuvers are executed, thereby managing computing resources more efficiently.
3Reliability
If the rule system executes actions based on machine learning model selections, then invalid behavior selections are prevented, but the processing time increases
Solution Approach 1:
The patent implements preliminary action by pre-defining the action mask with valid action criteria before the simulation runs. The action mask is configured in advance with the rules for acceptable maneuvers, allowing it to quickly validate and filter actions during execution without requiring complex real-time computations. This preliminary setup ensures only valid behaviors are executed while minimizing processing time during actual simulation operation.
Data Source
AI summary
A vehicle management system comprising a computer system and an agent. The agent comprises a machine learning model and a rule system. The machine learning model system is trained to receive observations for the vehicle system and select a behavior for the vehicle system in response to receiving the observations. The rule system is configured to select a set of actions to execute the behavior for the vehicle system in response to a selection of the behavior by the machine learning model system.