HD Map Generation With Sensor Fusion and Trajectory Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in generating accurate and consistent high-definition maps for navigation, which is crucial for safe and effective driving, as existing maps may lack detail and consistency between sensor scans, leading to inaccurate position estimation and increased risk of collisions.
Innovation Solution
A method and system that combine sensor data from different types, such as LiDAR and IMU, to generate a high-definition map by sorting data elements into structural categories, determining a mapping trajectory, and updating the map based on the vehicle's motion, using a dual-channel extended Kalman filter to align asynchronous input information and estimate the vehicle's state.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensor data from different types (LiDAR, IMU) are combined to generate high-definition maps, then mapping detail and accuracy are improved, but data processing complexity increases
Solution Approach 1:
The patent segments the complex sensor data processing into distinct modules: a dual-channel extended Kalman filter processes asynchronous sensor data (LiDAR, IMU, GPS) separately in parallel channels, while a structural data categorization system organizes mapping data into predefined categories. This segmentation reduces overall system complexity by breaking down the intricate task of fusing multiple sensor types into manageable, independent processing stages.
Solution Approach 2:
The patent transforms heterogeneous sensor data with different frequencies and formats into a unified parameter representation. The dual-channel extended Kalman filter converts asynchronous sensor inputs into synchronized state estimates, and the structural data categorization system standardizes mapping features into consistent parameter formats, enabling accurate fusion despite original parameter differences.
2Stability of the object's composition
If asynchronous sensor data is aligned to consistent time points, then map consistency is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary time alignment of asynchronous sensor data through the dual-channel extended Kalman filter, which predicts sensor states at consistent time points before actual data arrival. This preliminary action ensures that when data is processed, it is already temporally synchronized, eliminating the need for time-consuming post-processing alignment operations.
Solution Approach 2:
The extended Kalman filter implements continuous feedback by constantly comparing predicted sensor states with actual measurements and adjusting predictions accordingly. This feedback mechanism maintains temporal consistency across asynchronous sensors by dynamically correcting time alignment based on real-time performance, ensuring map consistency without exhaustive processing.
3Measurement precision
If high-definition maps are updated continuously based on vehicle motion, then navigation accuracy is improved, but computational load increases
Solution Approach 1:
The patent implements periodic map updates synchronized with vehicle motion events rather than continuous updates. The system triggers map regeneration based on predefined conditions such as distance traveled, time intervals, or significant positional changes detected by the extended Kalman filter. This periodic approach maintains navigation accuracy by updating maps at optimal intervals while significantly reducing computational load compared to continuous updating.
4Manufacturing precision
If structural data categorization is applied to sort data elements, then map detail is improved, but data processing complexity increases
Solution Approach 1:
The patent applies structural data categorization to organize mapping data into specific local categories such as road geometry, signage, pedestrians, and environmental features. Each category receives targeted processing appropriate to its characteristics, improving map detail by capturing relevant features with appropriate precision while avoiding unnecessary processing of unrelated data elements.
Data Source
AI summary
A method may include obtaining sensor data about a total measurable world around an autonomous vehicle. The sensor data may be captured by sensor units co-located with the autonomous vehicle. The method may include generating a mapping dataset including the obtained sensor data and identifying data elements that each represents a point in the mapping dataset. The method may include sorting the data elements according to a structural data categorization that is a template for a high-definition map of the total measurable world and determining a mapping trajectory of the autonomous vehicle. The mapping trajectory may describe a localization and a path of motion of the autonomous vehicle. The method may include generating the high-definition map based on the structural data categorization and relative to the mapping trajectory of the autonomous vehicle, and the high-definition map may be updated based on the path of motion of the autonomous vehicle.