IMU Pose Filtering With Learned Orientation Change Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing IMU-based 6DOF pose tracking systems suffer from accuracy drift due to sensor noise and bias, necessitating complex and costly sensor fusion with external measurements, while machine learning-based methods struggle to stabilize orientation change measurements without additional physical inputs.
Innovation Solution
Implementing a machine learning network to predict orientation change measurements using a sequence-to-sequence regression Transformer architecture, which provides learned orientation change feedback to an Extended Kalman Filter (EKF) for improved pose estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensor fusion with external measurements is used to correct IMU drift, then pose estimation accuracy is improved, but device complexity and cost increase
Solution Approach 1:
The patent replaces physical sensor fusion systems with a computational approach using machine learning models. Instead of fusing multiple physical sensors (cameras, lasers, ultrasonic sensors), the system uses a neural network to predict orientation changes from IMU data alone, substituting mechanical sensor fusion with intelligent signal processing.
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between IMU measurements and pose estimation. This intermediary (the neural network) processes IMU data and predicts orientation changes, effectively mediating the relationship between sensor input and output without requiring additional physical sensors.
2Device complexity
If machine learning-based methods are used to predict orientation changes, then reliance on external sensors is reduced, but computational processing requirements increase
Solution Approach 1:
The patent applies partial action by using a neural network that processes only the essential features from IMU data rather than attempting to process all possible sensor information. The model focuses computational resources on predicting orientation changes using a subset of relevant IMU measurements, avoiding unnecessary processing of redundant data.
3Use of energy by moving object
If traditional IMU integration is used without machine learning, then computational processing is simpler, but orientation change measurements are unstable and drift occurs
Solution Approach 1:
The patent implements feedback by using the neural network to continuously predict and correct orientation changes based on IMU measurements. The predicted orientation changes are fed back into the pose estimation system, creating a closed-loop correction mechanism that stabilizes measurements and reduces drift over time.
Data Source
AI summary
Systems and techniques are provided for determining a pose. A process can include obtaining inertial measurement unit (IMU) data from an IMU associated with a device. The IMU data can be used to determine a propagated state associated with a state estimation engine, wherein the propagated state includes an initial orientation estimate corresponding to a pose of the device. The state estimation engine can comprise an Extended Kalman Filter (EKF). A predicted orientation measurement can be generated using a first machine learning network to process the IMU data and the initial orientation estimate included in the propagated state associated with the state estimation engine. An updated state associated with the state estimation engine can be determined based on using the predicted orientation measurement to update the propagated state. A device pose estimate can be determined based on the updated state associated with the state estimation engine.