Aircraft Neural Network Training With Simulated Control Exploration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training an autonomous aircraft using data from human pilots is time-consuming and costly, and it may not cover all potential scenarios that the aircraft could encounter.
Innovation Solution
A system that uses a flight simulator to train a neural network by providing control inputs, generating state data, and updating weights based on reward values, allowing the neural network to learn from simulated scenarios, including those not commonly encountered by pilots.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training data is collected from human pilots, then the neural network learns from real-world experience, but the training process becomes time-consuming and costly
Solution Approach 1:
The patent creates a virtual copy of the aircraft environment through a flight simulator that replicates real-world flight dynamics and scenarios. This virtual environment allows the neural network to be trained on simulated pilot behaviors and flight conditions without requiring actual physical flight testing, thereby reducing training time and costs while maintaining data quality through realistic simulation models
2Reliability
If training data is collected from human pilots, then the neural network learns from existing expertise, but it cannot cover potential new scenarios that the aircraft could encounter
Solution Approach 1:
The flight simulator enables dynamic generation of training scenarios that can adapt to any conceivable flight condition, including rare or hypothetical situations. The simulator can dynamically adjust environmental parameters, aircraft states, and scenario complexity to cover the full range of possible operating conditions, allowing the neural network to learn responses to scenarios beyond those encountered by human pilots
Solution Approach 2:
The system performs preliminary action by pre-training the neural network on a comprehensive set of simulated scenarios before deployment. This includes pre-generating training data for edge cases and uncommon situations that may not be frequently encountered in real operations, enabling the aircraft to be prepared in advance for potential new scenarios
3Productivity
If a flight simulator is used for training, then resource consumption is reduced, but the complexity of the training system increases
Solution Approach 1:
The flight simulator acts as an intermediary layer between the neural network training process and the physical aircraft. This intermediary virtual environment handles the complexity of flight dynamics, sensor simulations, and scenario generation, allowing the neural network to be trained efficiently without direct complexity in the training infrastructure itself
Data Source
AI summary
A device includes a control input generator and a neural network trainer. A flight simulator is configured to generate first state data responsive to a first control input from the control input generator and to provide the first state data to a first neural network to generate a candidate second control input. The control input generator is also configured to select, based on a random value, a second control input from between the candidate second control input and a randomized offset control input that is based on a random offset applied to the first control input. The flight simulator is configured to generate second state data responsive to the second control input from the control input generator. The neural network trainer is configured to update weights of the first neural network based, at least in part, on the first state data and the second state data.