Observability-Constrained VINS Estimator Consistency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Vision-aided inertial navigation systems (VINS) face inconsistency issues due to spurious information gain along unobservable directions, leading to larger estimation errors and divergence, particularly when using linearized estimators like the Extended Kalman Filter (EKF).
Innovation Solution
The implementation of an Observability-Constrained VINS (OC-VINS) approach, which enforces unobservable directions to prevent spurious information gain by modifying the state transition and measurement models to ensure consistency, specifically by maintaining the nullspace structure and updating Jacobians to adhere to the true system's observability properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If linearized estimators like Extended Kalman Filter are used in VINS, then computational complexity is reduced and real-time processing is enabled, but estimator inconsistency occurs due to spurious information gain along unobservable directions
Solution Approach 1:
The patent modifies the state transition and measurement models by changing the parameter representation to explicitly account for unobservable directions. This involves transforming the system matrices to separate observable and unobservable subspaces, allowing the linearized estimator to maintain consistency by preventing information leakage along unobservable directions while preserving real-time processing capability
Solution Approach 2:
The patent introduces an intermediary transformation matrix that acts as a mediator between the standard linearized estimator and the system state. This intermediary structure enforces the unobservable direction constraints without requiring complex nonlinear optimization, thus maintaining computational efficiency while improving estimator consistency
2Adaptability or versatility
If calibration parameters are estimated using external calibration targets not aligned with gravity, then calibration versatility is improved, but measurement precision deteriorates due to coupling between calibration parameters and attitude estimates
Solution Approach 1:
The patent segments the estimation problem by separating calibration parameter estimation from attitude estimation through nullspace projection. This segmentation allows independent optimization of each parameter set, preventing error coupling and improving measurement precision while maintaining the ability to use arbitrarily oriented calibration targets
Solution Approach 2:
The patent transforms the calibration estimation problem by adding a new dimension of analysis through nullspace decomposition. This dimensional transformation separates the estimation space into observable and unobservable subspaces, allowing precise calibration parameter estimation even when calibration targets are not aligned with gravity, thus improving both versatility and precision
Data Source
AI summary
This disclosure describes various techniques for use within a vision-aided inertial navigation system (VINS). A VINS comprises an image source to produce image data comprising a plurality of images, and an inertial measurement unit (IMU) to produce IMU data indicative of a motion of the vision-aided inertial navigation system while producing the image data, wherein the image data captures features of an external calibration target that is not aligned with gravity. The VINS further includes a processing unit comprising an estimator that processes the IMU data and the image data to compute calibration parameters for the VINS concurrently with computation of a roll and pitch of the calibration target, wherein the calibration parameters define relative positions and orientations of the IMU and the image source of the vision-aided inertial navigation system.