Cooperative UAS Pose Graph Navigation in GPS-Denied Flight
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fixed-wing unmanned aircraft systems (UASs) face challenges in navigation when absolute location information, such as GPS, is unavailable, partially unavailable, or degraded.
Innovation Solution
The system employs a front-end nodule to estimate relative pose changes using an extended Kalman filter, which periodically publishes these estimates to a back-end module for global pose model generation and optimization. Additionally, UASs share range information to improve back-end pose graph optimization through coordinated resets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If UAS relies solely on front-end relative pose estimation without back-end optimization, then computational complexity is reduced, but navigation accuracy deteriorates due to error accumulation
Solution Approach 1:
The navigation system is divided into two independent modules: front-end relative pose estimation and back-end global pose optimization. The front-end uses extended Kalman filter for real-time relative pose estimation with lower computational complexity, while the back-end performs periodic pose graph optimization to correct accumulated errors and improve navigation accuracy. This segmentation allows each module to operate at appropriate computational levels.
Solution Approach 2:
The back-end pose graph optimization performs preliminary correction of accumulated errors before they significantly degrade navigation accuracy. By periodically optimizing the global pose model using previously estimated relative poses, the system proactively prevents error accumulation from compromising navigation performance.
2Device complexity
If UAS operates independently without sharing range information, then system simplicity is maintained, but pose graph optimization accuracy deteriorates
Solution Approach 1:
Multiple UASs merge their individual pose graphs by sharing range information and performing coordinated resets. The range measurements between UASs serve as additional constraints that link their respective pose graphs, enabling more accurate global pose estimation through cooperative optimization while maintaining individual system simplicity.
Solution Approach 2:
UASs exchange range information that provides feedback constraints for pose graph optimization. This feedback mechanism allows each UAS to improve its pose estimation accuracy by incorporating measurements from other UASs, enhancing the overall optimization accuracy without requiring complex centralized control.
3Measurement precision
If front-end reset frequency is increased, then error accumulation is reduced, but information loss increases due to frequent covariance zeroing
Solution Approach 1:
The back-end pose graph optimization performs preliminary correction of pose errors before frequent resets would cause significant information loss. By periodically optimizing the global pose model, the system maintains accuracy while allowing the front-end to reset at appropriate intervals without excessively zeroing out valuable covariance information.
Solution Approach 2:
The back-end pose graph optimization acts as an intermediary that preserves navigation information across front-end resets. When the front-end resets and zeros its covariance, the back-end maintains the historical pose graph information, preventing permanent information loss and enabling continuous improvement of navigation accuracy.
Data Source
AI summary
A UAS may use camera-captured ground feature data to iteratively estimate pose change through a Kalman filter. To avoid excessive drift, the UAS may periodically perform a reset by publishing delta pose and covariance as an edge in a global pose graph, by zeroing out the delta pose and covariance. To further improve global pose estimation, two UASs may share range information and associated pose graph information, thereby providing an additional constraint for each UAS to use in its pose graph optimization.