Vehicle Localization Error Prediction Using Map-Aware Sensor Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional GPS-based localization systems in autonomous driving face significant accuracy challenges due to systematic errors in urban environments, such as multipath reflections, which can result in location errors of up to 30 meters, making it difficult to ensure safe and efficient route planning.
Innovation Solution
A machine learning model is employed to predict sensor errors by processing data from various vehicle sensors and map data, reducing the localization search space and enabling faster and more accurate vehicle positioning by using predicted sensor error priors in sensor fusion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional GPS-based localization is used in urban environments, then the system is simple and low-cost, but the localization accuracy deteriorates significantly due to multipath reflections and other sensor errors
Solution Approach 1:
The system performs preliminary characterization of sensor errors by collecting GPS data and comparing it with ground truth locations during a mapping phase. Error patterns are identified and stored as correction data before actual localization occurs, enabling the system to compensate for known error sources without adding complexity to real-time operation
Solution Approach 2:
The patent introduces map data and environmental features as an intermediary between the GPS sensor and the final location estimate. By using map-matching and comparing GPS readings with expected locations based on map data and sensor fusion, the system mediates the inaccurate GPS measurements to produce accurate localization
2Measurement precision
If GPS error characterization and correction is implemented, then localization accuracy improves, but the processing time and computational resources increase
Solution Approach 1:
Error characterization data is collected and processed in advance during a mapping phase, storing pre-computed correction information in map data structures. This eliminates the need for real-time error analysis during localization, reducing processing time to simple lookups and adjustments
Solution Approach 2:
The system applies selective correction based on detected error patterns rather than processing all possible corrections. When GPS error is detected through sensor fusion inconsistencies, the system applies targeted map-matching and error compensation only for the affected components, avoiding unnecessary full-system reprocessing
3Speed
If sensor error priors are used in sensor fusion, then the convergence speed of localization improves, but the complexity of the fusion algorithm increases
Solution Approach 1:
Error priors are pre-computed and stored in the map data structures during the mapping phase. The priors include expected error characteristics for different locations and environmental conditions, allowing the localization algorithm to apply predetermined corrections without performing complex real-time error analysis
Solution Approach 2:
The system uses feedback from multiple sensors (odometry, map matching, GPS) to detect inconsistencies that indicate GPS error. When error is detected, the system adjusts the weighting of GPS data in the fusion algorithm based on pre-characterized error patterns, creating a feedback loop that automatically adapts to error conditions without manual intervention
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An approach is provided for providing predictive classification of location sensor error for vehicle localisation. The approach involves, for example, receiving sensor data from at least one sensor (105), the sensor data collected at a geographic location. The approach also involves extracting a set of input features from the sensor data, map data (113) representing the geographic location, or combination thereof. The approach further involves processing the set of input features using a machine learning model (115) to calculate a predicted sensor error of a target location sensor operating at the geographic location. The machine learning model (115), for instance, has been trained on ground truth sensor error data to use the set of input features to calculate the predicted sensor error.