Aircraft Dynamics Prediction Using Residual Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional physics-based methods for predicting aircraft trajectories require a pre-existing model of the exact aircraft and may need unavailable state variables, making them unsuitable for autonomous applications, and existing learning-based methods fail to accurately model aircraft dynamics.
Innovation Solution
A neural network with a feedforward skip connection is used to predict aircraft trajectories, employing a residual output method and loss function optimization to handle angular variables and compressive encoding for state representation, enabling accurate predictions in adversarial and cooperative scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physics-based dynamic equations are used to predict aircraft trajectory, then prediction accuracy is improved, but the method requires a pre-existing model of the exact aircraft and unavailable state variables, making it unsuitable for autonomous applications
Solution Approach 1:
The patent creates a simplified virtual model (copy) of the aircraft dynamics that mimics the essential behavior without requiring the complete physical model. This virtual model uses only the state variables actually available to the autonomous system, eliminating the need for unavailable parameters while preserving predictive accuracy for autonomous decision-making
Solution Approach 2:
The patent replaces the traditional physics-based mechanical modeling approach with a data-driven virtual model that learns aircraft dynamics from operational data. This substitution eliminates the requirement for detailed physical models and unavailable state variables, enabling autonomous systems to predict trajectories using only sensor data actually available during operation
2Ease of operation
If traditional learning-based methods are used to predict aircraft trajectory, then applicability to autonomous systems is improved, but prediction accuracy deteriorates due to inability to accurately model aircraft dynamics
Solution Approach 1:
The patent transforms the learning approach by changing from traditional supervised learning to a novel virtual model learning framework that optimizes for dynamics prediction. The loss function is specifically designed to penalize errors in predicted state transitions, enabling accurate modeling of aircraft dynamics while maintaining autonomy applicability
Solution Approach 2:
The patent creates a virtual copy of the aircraft dynamics system that learns to replicate the true dynamics from operational data. This virtual model accurately captures the complex nonlinear behavior of aircraft while using only available state variables, achieving both accuracy and autonomous system compatibility
3Adaptability or versatility
If a virtual model of aircraft dynamics is learned from operational data, then applicability to autonomous systems without pre-existing models is improved, but model complexity increases
Solution Approach 1:
The patent extracts only the essential dynamics information needed for autonomous decision-making from the complex aircraft operational data. By focusing solely on learning the state transition dynamics rather than modeling all aircraft characteristics, the virtual model achieves adaptability without excessive complexity
Solution Approach 2:
The patent segments the learning task into specifically predicting state transitions rather than modeling the entire aircraft system. This segmentation of the dynamics prediction task from complete system modeling reduces computational complexity while maintaining the adaptability needed for autonomous applications
Data Source
AI summary
A computer-implemented method for predicting behavior of aircraft is provided. The method comprises inputting a current state of a number of aircraft into a number of hidden layers of a neural network, wherein the neural network is fully connected. An action applied to the aircraft is input into the hidden layers concurrently with the current state. The hidden layers, according to the current state and current action, determine a residual output that comprises an incremental difference in the state of the aircraft resulting from the current action. A skip connection feeds forward the current state of the aircraft, and the residual output is added to the current state to determine a next state of the aircraft.