IMU Signal Correction Using Machine Learning for Position Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing IMU calibration methods fail to account for sensor distortion, sensitivity differences, white noise, and inaccuracies in determining device motion, leading to errors in position estimation.
Innovation Solution
An information processing apparatus utilizing machine learning models to estimate bias errors and scale factor errors in IMU data, followed by a neural network to correct and integrate acceleration data for precise position estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional IMU calibration methods are used to correct bias error, scale factor error, and non-orthogonal error, then partial signal accuracy is improved, but sensor distortion errors and sensitivity difference errors cannot be calibrated
Solution Approach 1:
The patent transforms the calibration approach by changing from correcting only static parameters (bias, scale factor, non-orthogonal error) to dynamically estimating and correcting time-varying parameters including sensor distortion and sensitivity differences. The machine learning model continuously adapts calibration parameters based on actual device motion states, enabling comprehensive error correction that was previously unachievable with conventional static calibration methods.
2Ease of operation
If the device is determined to be stationary based on successive identical IMU readings, then calibration can proceed, but the device may be incorrectly determined as stationary when moving at low speed or just starting to move
Solution Approach 1:
The patent introduces machine learning models as intermediaries between raw IMU readings and motion state determination. Instead of directly comparing successive readings, the system uses trained models to analyze patterns in the data and accurately distinguish between true stationary states and low-speed motion, thereby eliminating false positives in stationary detection while maintaining operational simplicity.
3Productivity
If conventional calibration methods are applied, then processing time is reduced, but position estimation error accumulates over long periods due to uncorrected white noise and sensor errors
Solution Approach 1:
The patent replaces conventional mathematical calibration algorithms with machine learning-based error estimation systems. The ML models, trained on comprehensive error characteristics including white noise patterns, provide more accurate real-time corrections that prevent error accumulation during long-term integration, thereby maintaining both processing efficiency and long-term accuracy.
Data Source
AI summary
An information processing apparatus 1 sequentially receives a plurality of input data between predetermined time TS and time TE, and uses model information to estimate predetermined estimated data at a time point of time T (TS<T<TE) based on the received input data, the model information having been trained by machine learning so as to estimate the estimated data at the time point of the time T on the basis of the received input data.