IMU Drift Correction via Neural Network Vector Offsets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion data gathering techniques suffer from drift issues, leading to inaccurate data, particularly in environments with vibrations, which can result in misdiagnosis, low-quality physical therapy, or improper training, due to the limited self-correction abilities of single-device systems.
Innovation Solution
A system comprising a processor and memory that receives data from motion sensors and cameras, applies machine learning models, specifically neural networks, to generate vectors for offsetting sensor data, improving data accuracy by mitigating drift and enhancing inertial odometry measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion data is gathered from a single device with limited self-correction abilities, then device complexity is reduced, but measurement precision deteriorates due to drift
Solution Approach 1:
A machine learning model serves as an intermediary between the motion sensor and the final measurement output. The model receives raw sensor data, processes it to identify and correct drift patterns, and generates corrected motion data. This intermediary layer enables a single device to achieve high measurement precision without requiring complex multi-device configurations.
Solution Approach 2:
The motion sensor system performs self-correction of drift errors through integrated machine learning processing. The system automatically identifies drift patterns in its own measurements and applies corrections without external intervention or additional reference devices, enabling a single device to maintain high measurement precision over time.
2Measurement precision
If machine learning models are used to correct sensor data drift, then measurement precision is improved, but device complexity increases due to additional processing requirements
Solution Approach 1:
The system applies machine learning processing selectively to correct specific drift patterns rather than processing all sensor data equally. The model identifies and corrects only the portions of data affected by drift, maintaining measurement precision while minimizing unnecessary processing complexity.
3Measurement precision
If drift correction is applied in real-time, then measurement precision is maintained over time, but use of energy increases due to continuous processing
Solution Approach 1:
The machine learning model performs drift correction at periodic intervals rather than continuously processing every data point. This approach maintains measurement precision over time by regularly updating corrections while reducing energy consumption compared to continuous real-time processing of all sensor data.
Data Source
AI summary
This disclosure relates to systems, media, and methods for mitigating measurement drift and improving IMU odometry measurement. In an embodiment, the system may perform operations including receiving first sensor data from at least one motion sensor; receiving 3-dimensional (3-D) motion data based on motion detected by at least one camera; inputting model input data into a machine learning model configured to generate at least one vector, the model input data being based on the received first sensor data and the received 3-D motion data; and apply the at least one vector as an offset.