UAV Autopilot Learning Model Weighting for Unstable Flight Conditions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Unmanned aerial vehicles (UAVs) face challenges in achieving accurate and stable automatic piloting due to varying environments and conditions, which affect the learning process of the relationship between piloting and vehicle behavior.
Innovation Solution
An information processing apparatus that includes a learning unit and a determination unit, which adjusts the weight given to the relationship between piloting and vehicle behavior based on conditions such as low-visibility flights, anomaly-affected flights, signal-missing flights, and flights with excessive or unexpected behavior, to focus learning on suitable environments and conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If machine learning is performed on all flight data including low-visibility flights, anomaly-affected flights, and signal-missing flights, then the quantity of learning data increases, but the accuracy and stability of automatic piloting deteriorates due to unsuitable learning conditions
Solution Approach 1:
The patent segments flight data into suitable and unsuitable learning data based on flight conditions. The determination unit classifies flights into categories (low-visibility, anomaly-affected, signal-missing, normal) and the learning unit processes only suitable data, separating beneficial learning material from harmful noise to improve model reliability.
Solution Approach 2:
The patent applies different quality standards to different portions of learning data. By assigning weight values where suitable learning data receives weight of 1 and unsuitable data receives weight of 0 or less, the system ensures high-quality training material is prioritized, improving overall learning accuracy.
2Adaptability or versatility
If all flight data is used for learning without discrimination, then the learning process covers more environments and conditions, but the learning efficiency and model quality deteriorates due to noise from unsuitable data
Solution Approach 1:
The determination unit performs preliminary classification of flight data before the learning unit processes it. By pre-identifying suitable learning data based on flight conditions, the system prepares high-quality training material in advance, preventing inefficient learning from unsuitable data.
Solution Approach 2:
The patent dynamically adjusts the weight given to different learning data based on flight conditions. The learning unit flexibly modifies learning weights during the training process, increasing weights for suitable data and decreasing or eliminating weights for unsuitable data, optimizing learning efficiency adaptively.
Data Source
AI summary
A server device performs machine learning on the relationship between the content of piloting of an aerial vehicle and the behavior of the aerial vehicle in response to the content of the piloting, and generates a learning model for automatically piloting the aerial vehicle. However, the aerial vehicle is piloted in various environments and conditions, and there are environments and conditions that are unsuited for achieving highly accurate and stable automatic piloting. Therefore, the server device performs the machine learning only in an environment or a condition suited for realizing the automatic piloting.