Neural INS Positioning for Vehicles in Weak GNSS Environments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle positioning systems face challenges in maintaining accurate positioning in weak GNSS environments without increasing hardware complexity or cost, as GNSS signals are obstructed in scenarios like tunnels and urban canyons, and INS systems suffer from accumulated errors.
Innovation Solution
A method involving training a position prediction network and a position correction network using INS data to predict and correct vehicle position, utilizing INS data as inputs, without adding additional sensors, by leveraging the difference between GNSS and INS measurements in two stages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If more vehicle-mounted sensors are added for auxiliary positioning, then positioning accuracy in weak GNSS environment is improved, but hardware cost and system complexity increase
Solution Approach 1:
The patent creates a virtual copy of GNSS positioning capability through neural network models (position prediction network and position correction network) that simulate GNSS position estimation using only INS data, eliminating the need for additional physical sensors while maintaining positioning accuracy
Solution Approach 2:
The patent replaces the mechanical approach of adding more physical sensors with an information processing approach using trained neural networks to predict and correct INS positioning errors, substituting hardware expansion with software-based intelligence
2Measurement precision
If GNSS/INS integrated navigation system is used, then positioning performance is improved to a certain extent, but positioning requirements in complex environments are still difficult to meet
Solution Approach 1:
The patent implements a feedback mechanism where the position correction network uses the difference between predicted position and actual GNSS position as correction information, continuously optimizing the INS positioning output by feeding back error compensation signals to improve reliability
Solution Approach 2:
The patent performs preliminary training of the position prediction and correction networks using normal GNSS data before deploying them in weak signal environments, preparing the models in advance with accurate training samples to ensure reliable performance when GNSS signals are unavailable
Data Source
AI summary
The present disclosure provides a vehicle positioning method and system in a weak GNSS environment. The method includes: preprocessing system operation data information of an autonomous vehicle, where the system operation data information includes latitude and longitude data acquired from the GNSS system and three-axis acceleration, three-axis angular velocity and heading information acquired from an INS system; training a position prediction network with the preprocessed system operation data information acquired from the INS system to finally output supervision information of the prediction network; training a position correction network according to the supervision information in step S2 and the preprocessed system operation data information and the supervision information of the position correction network; and finally outputting a predicted value for correction. In the weak GNSS environment or in case of GNSS interruption, the final predicted value is outputted through the position prediction network and the position correction network.