Deep Learning IMU Error Correction for Attitude Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low-priced inertial measurement unit (IMU) sensors suffer from large errors due to microelectromechanical systems (MEMS) methods, leading to inaccurate attitude estimation, especially in applications requiring precise angle determination like autonomous driving and robotics.
Innovation Solution
An apparatus and method utilizing deep learning to improve the resolution and attitude estimation accuracy of IMU sensors by incorporating data from higher-performance IMU sensors and additional sensors like camera, LiDAR, and encoder sensors, employing preprocessing and training with convolutional neural networks (CNN) or other deep learning architectures to enhance data processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If low-priced IMU sensors using MEMS method are used, then cost is reduced, but measurement precision deteriorates due to large errors
Solution Approach 1:
A deep learning-based performance improvement module is introduced as an intermediary between the low-priced IMU sensor and the final attitude estimation output. This module processes the raw sensor data through preprocessing and deep learning operations to correct errors and improve measurement precision without requiring a more expensive sensor
Solution Approach 2:
The patent replaces the need for high-precision mechanical MEMS sensors with a software-based deep learning system. Instead of relying on expensive hardware with superior physical characteristics, the solution uses neural networks to compensate for hardware limitations and achieve comparable or superior measurement accuracy
2Measurement precision
If deep learning processing is applied to improve IMU sensor accuracy, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The deep learning processing system is divided into distinct functional modules: a preprocessing module that prepares sensor data, a performance improvement module that applies deep learning operations, and an output module that generates improved attitude estimation. This segmentation allows each module to be optimized independently and simplifies the overall system architecture
Solution Approach 2:
The preprocessing module performs necessary data preparation and conditioning before the main deep learning processing occurs. This preliminary action includes normalizing sensor data, synchronizing multiple sensor inputs, and preparing the data in a format suitable for the neural network, thereby simplifying the main processing stage
Data Source
AI summary
The present invention relates to an apparatus and method for improving resolution and attitude estimation accuracy of an inertial measurement unit (IMU) sensor using deep learning. The apparatus for improving the resolution and the attitude estimation accuracy of the IMU sensor using deep learning according to the present invention includes an input unit configured to acquire data of an IMU sensor, a memory configured to store a performance improvement program of the IMU sensor, and a processor configured to execute the program, wherein the processor serves to improve accuracy and resolution of acceleration and angular velocity of the data of the IMU sensor by performing deep learning.