Inertial Sensor Localization Using Deep Learning Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current localization technologies, including GNSS, struggle with providing high-accuracy positioning both indoors and outdoors, especially for autonomous vehicles, due to limitations in sensor accuracy and availability, such as in areas with poor signal reception.
Innovation Solution
A deep learning neural network (DL NN) model is trained using inertial measurement unit (IMU) data, including angular velocities and accelerations, to determine a vehicle's location by calculating unique kinematic signatures and associating them with positioning measurements from other sources, enabling high-accuracy localization in real-time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensor fusion schemes are used for indoor localization, then localization accuracy is improved, but device complexity and cost increase
Solution Approach 1:
The patent extracts and utilizes only the necessary inertial sensor data (accelerometer and gyroscope measurements) for localization, eliminating the need for complex multi-sensor fusion schemes. By focusing on IMU data alone and processing it through neural networks, the system achieves accurate indoor localization without the complexity of integrating multiple sensor types.
Solution Approach 2:
The patent replaces complex mechanical sensor fusion systems with a computational approach using neural networks that process IMU data. This substitution transforms the localization problem from a hardware-intensive sensor fusion task into a software-based pattern recognition task, reducing device complexity while maintaining accuracy.
2Measurement precision
If multiple sensors (camera, LIDAR, inertial sensors) are used for high-accuracy outdoor localization, then localization accuracy is improved, but cost increases
Solution Approach 1:
The patent makes the IMU serve multiple functions: it provides both orientation data (through gyroscope) and position data (through accelerometer integration). This multi-functionality eliminates the need for separate specialized sensors, reducing the total number of sensors required while maintaining high localization accuracy through the neural network's ability to process and integrate these multi-purpose measurements.
3Adaptability or versatility
If GNSS is used for outdoor localization, then coverage area is improved, but measurement precision deteriorates in urban canyons and indoor environments
Solution Approach 1:
The patent introduces inertial sensors as an intermediary system that works independently of GNSS signal availability. The IMU provides continuous localization capability in environments where GNSS fails (urban canyons, indoors) by measuring motion directly, thus mediating between the need for wide coverage and the need for consistent high accuracy regardless of location.
Data Source
AI summary
A system and method for providing localization, including, during a training phase: obtaining a training dataset of accelerations, angular velocities, and known locations over time of vehicles moving in a defined area; and training a machine learning model to provide location estimation in the defined area based on the accelerations and angular velocities using the training dataset; and during runtime phase: obtaining runtime accelerations and angular velocities overtime of a vehicle moving in the defined area; and using the trained model to obtain current location of the vehicle based on the runtime acceleration and angular velocities.