Target Drone Wind Field Control for Real-Time Trajectory Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the wind field in a geographical area of interest are cumbersome and require significant computing time, making them unsuitable for real-time execution on the onboard computer of a target drone.
Innovation Solution
A method involving a system with measuring devices, a ground station, and a target drone that uses a combination of stationary and mobile measuring devices to collect wind data, which is then processed to estimate the wind field quickly and efficiently, allowing for nearly real-time trajectory planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If precise simulation software using airflow physical equations is used to determine the wind field, then measurement precision is improved, but device complexity increases and computing time becomes excessive for real-time execution
Solution Approach 1:
The patent creates a simplified copy of the complex simulation model by training a neural network on simulation data. The neural network replicates the wind field estimation functionality of the complex simulation software but with dramatically reduced computational requirements, enabling real-time execution on embedded systems while maintaining acceptable accuracy.
Solution Approach 2:
The patent replaces the mechanical/computational simulation system (solving airflow physical equations) with an intelligent system (neural network). This substitution transforms the complex numerical computation into a lightweight machine learning inference process that can run in real-time on embedded hardware.
2Measurement precision
If precise simulation software is used to determine the wind field, then measurement precision is improved, but loss of time increases due to significant computing time required for convergence
Solution Approach 1:
The patent performs the computationally intensive work in advance by generating simulation data and training the neural network offline. Once trained, the model can provide real-time wind field estimates without requiring complex computations during actual operation, thus eliminating the time loss during critical flight operations.
Solution Approach 2:
The neural network serves as a pre-trained copy of the simulation model that can rapidly reproduce wind field estimates without re-solving the complex physical equations, enabling real-time performance while maintaining the accuracy benefits of the original simulation approach.
3Productivity
If lightweight embedded software is used for real-time execution, then device complexity is reduced and computing speed is improved, but measurement precision deteriorates compared to precise simulation software
Solution Approach 1:
The patent creates a trained neural network model that copies the essential wind field estimation capabilities of the complex simulation software. This copy runs efficiently on embedded systems while maintaining sufficient accuracy for real-time trajectory optimization, bridging the gap between precision and speed.
Solution Approach 2:
The patent transforms the problem from solving complex physical equations with many parameters to using a neural network with optimized parameters. By changing the computational approach and adjusting network parameters during training, the system achieves real-time performance with acceptable accuracy for the application.
4Measurement precision
If complex simulation software is used, then measurement precision is improved, but ease of operation deteriorates due to inability to execute on onboard computer
Solution Approach 1:
The patent replaces the complex simulation software with a neural network system that is natively suitable for embedded deployment. This substitution maintains the ability to provide accurate wind field estimates while making the system operable on resource-constrained onboard computers, thus improving ease of operation.
Solution Approach 2:
The neural network serves as a portable copy of the simulation model that can be deployed on various embedded platforms. This copy eliminates the dependency on complex software environments and high-performance computing resources, enabling straightforward deployment on onboard computers.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Method (100) for controlling a motion of a target drone (4) inside a geographical area of interest (Z), comprising the steps of : - acquisition (110), by a plurality of measuring devices located throughout the geographical area of interest, of a plurality of measures of values of wind parameters, the wind parameters comprising an amplitude and an orientation of the wind; - determining (130), from the plurality of measures, boundary conditions, defined as values of the wind parameters at a boundary of a domain surrounding the geographical area of interest; - determining (140), by using a first parametric function that depends on the boundary conditions only, a wind field, the wind field providing the values of the wind parameters in any point inside the domain; and, - controlling (160) the target drone by taking into account the wind field.