Aerial Vehicle Visual Localization With Dynamic Uncertainty Estimation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing GPS-denied localization methods for aerial vehicles face challenges in accurately matching aerial images to satellite images due to varying noise levels and lack of reliable uncertainty estimation, leading to inaccurate localization results, especially in real-time applications.

Innovation Solution

A system combining an evidential deep learning neural network with a recursive Bayesian estimator, such as a particle filter, provides dynamic uncertainty estimates by comparing real-time aerial images to stored satellite imagery, using a single neural network pass to predict both similarity and uncertainty, thereby improving localization accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If hand-tuned measurement models are used to provide uncertainty estimates, then statistical estimators can process images over time to combat sensor noise, but the models are difficult to calibrate and provide only static uncertainty estimates that flatten differences in visual measurement noise levels

Engineering Contradiction:
Improvelocalization accuracyVSAvoidmodel calibration complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The neural network performs self-calibration by automatically learning the relationship between image features and uncertainty during training on labeled data, eliminating the need for manual calibration of measurement models. The system serves itself by using training data to automatically adjust uncertainty estimates without human intervention.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system transitions from static uncertainty parameters to dynamic uncertainty parameters that are learned as neural network outputs. The uncertainty estimates are no longer fixed hand-tuned values but are instead computed parameters that adapt to the specific characteristics of each image based on learned patterns from training data.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If sampling methods are used to estimate uncertainty with machine learning systems, then uncertainty estimates can be obtained, but the approaches are computationally expensive and consume excessive power and processing resources

Engineering Contradiction:
Improveuncertainty estimation accuracyVSAvoidcomputational energy consumption
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system extracts uncertainty estimation capability from computationally intensive sampling processes and integrates it directly into the neural network's forward pass. By removing the sampling step, the system obtains uncertainty estimates through a single efficient neural network execution rather than through multiple expensive sampling operations.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent replaces the mechanical sampling process (repeated passes through the network or ensemble of networks) with a neural network-based uncertainty estimation mechanism. This substitution uses learned parameters from training data to directly compute uncertainty without the iterative sampling mechanism, significantly reducing computational requirements.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Device complexity

If static measurement models are used, then the system is simpler to implement, but the models assume the same level of uncertainty for all input measurements and produce inaccurate sub-optimal results when visual measurements have variable noise levels

Engineering Contradiction:
Improvemeasurement model simplicityVSAvoidlocalization precision
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The system transitions from static measurement models to dynamic models where uncertainty estimates vary based on the specific characteristics of each image. The neural network learns to assign different uncertainty levels to different images based on their visual characteristics, enabling adaptive uncertainty estimation that improves localization precision.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes from fixed uncertainty parameters to learned parameters that adapt to input data characteristics. The neural network automatically adjusts uncertainty estimates based on the specific image being processed, allowing the system to handle variable noise levels in visual measurements accurately.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4707735A1Visual location of aerial vehicles using dynamic aleatoric uncertainty
Publication Date: 2026.03.11 THE BOEING CO
  • EP4707735A1 patent drawingFigure 1
  • EP4707735A1 patent drawingFigure 2
  • EP4707735A1 patent drawing

AI summary

Techniques for localizing a vehicle in real time using dynamic uncertainty estimates are presented. The techniques include obtaining a terrain image captured by the vehicle; passing the terrain image to a trained evidential deep learning neural network subsystem, from which a dynamic uncertainty value and a first feature vector are obtained in real time; for each of a plurality of candidate terrain locations, comparing the first feature vector to a respective second feature vector representative of a candidate terrain location, from which a respective similarity score is obtained; for at least one of the plurality of candidate terrain locations, updating in real time, by a recursive Bayesian estimator, a respective location weight based on the dynamic uncertainty value and the respective similarity score; estimating, in real time, a location of the vehicle based on the plurality of location weights; and providing the location of the vehicle.