DRL-RNP: Deep Reinforcement Learning-Based Optimized RNP Flight Procedure Execution.

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Control Systems

Background:

• Performance-based navigation (PBN) and Required Navigation Performance (RNP) procedures enhance airspace efficiency and reduce reliance on ground facilities.
• Aircraft approach phases are critical and complex, demanding precise navigation.
• Existing methods may not fully optimize fuel consumption or adapt to dynamic conditions like wind.

Purpose of the Study:

• To develop and evaluate a deep reinforcement learning (DRL)-based system for executing RNP procedures (DRL-RNP).
• To explore minimum fuel consumption flight paths during RNP approaches under windy conditions.
• To ensure compliance with RNP safety specifications, including protection areas and obstruction clearance.

Main Methods:

• Implementation of a DRL algorithm to control a six degrees of freedom fixed-wing aircraft model.
• Simulation of RNP approach procedures with a focus on dynamic environmental factors (wind).
• Utilizing a reward system within the DRL framework to incentivize fuel efficiency and safety compliance.

Main Results:

• The DRL-controlled aircraft successfully completed simulated RNP approach procedures.
• The system met all specified safety criteria, including protection areas and obstruction clearance altitude.
• DRL effectively identified flight paths that minimized fuel consumption under windy conditions.

Conclusions:

• Deep reinforcement learning is a viable method for executing complex RNP procedures.
• DRL-RNP offers a promising approach for optimizing fuel efficiency in aviation.
• The DRL method can be applied to simulate aircraft navigation, verify RNP procedures, and aid in their evaluation.