Dual agent reinforcement learning based system for autonomous operation of aircraft
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Pilots may not always operate aircraft in the most efficient or optimized manner, leading to suboptimal fuel consumption and performance, especially during emergency situations where stress and lack of real-time cross-validation can impact decision-making.
Innovation Solution
A dual agent reinforcement learning autonomous system (DARLAS) is developed to assist pilots by computing optimal velocities, altitudes, and headings for aircraft, using a safe agent to prioritize safety and a cost agent to optimize fuel efficiency, with the system trained in both simulated and automated scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If pilots operate aircraft manually, then they can handle emergency situations, but fuel consumption and performance are not optimized
Solution Approach 1:
The autonomous system is divided into two separate agents: a safe agent that prioritizes safety and emergency handling, and a cost agent that optimizes fuel consumption and performance. This segmentation allows each agent to specialize in its function while working together through the neural network to achieve both safety and efficiency goals that neither could accomplish alone.
2Reliability
If pilots follow checklists and protocols, then they can handle emergencies, but the actions are not tied to current location and system state
Solution Approach 1:
The system continuously receives feedback from the aircraft's current state (location, system status, emergency conditions) and uses this real-time information to dynamically adjust its recommendations. The neural network processes current state data to generate context-appropriate actions, ensuring that safety guidance is always adapted to the specific situation rather than following fixed protocols.
3Productivity
If pilots make decisions without real-time cross-validation, then they maintain decision speed, but outcomes may not be optimal
Solution Approach 1:
The dual-agent system acts as an intermediary that provides real-time cross-validation of pilot decisions. The safe agent and cost agent independently evaluate potential actions and provide complementary perspectives - one focused on safety constraints and the other on efficiency optimization. This intermediary evaluation occurs rapidly without replacing pilot authority, maintaining decision speed while improving accuracy through combined expert assessment.
Data Source
AI summary
A dual agent reinforcement learning autonomous system (DARLAS) for the autonomous operation of aircraft and/or provide pilot assistance. DARLAS includes an artificial neural network, safe agent, and cost agent. The safe agent is configured to calculate safe reward Q values associated with landing the aircraft at a predetermined destination or calculated emergency destination. The cost agent is configured to calculate cost reward Q values associated with maximum fuel efficiency and aircraft performance. The safe and cost reward Q values are based on state-action vectors associated with an aircraft, which may include state data and action data. The system may include a user output device that provides an indication of an action to a user. The action corresponds to an agent action having the highest safe reward Q value and the highest cost require Q value. DARLAS prioritizes the highest safe reward Q value in the event of conflict.