Lane Uncertainty Modeling with Recursive Least Squares
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle systems face challenges in accurately modeling and tracking lane lines due to uncertainty in observed points, which can affect semi-autonomous and autonomous operations by not adequately accounting for the precision of lane models.
Innovation Solution
A method and system that utilize sensors such as cameras, lidar, and radar to obtain observation points and their corresponding uncertainty values, generating and updating both lane models and uncertainty models using recursive least squares (RLS) adaptive filters to represent the path and precision of lane lines, respectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If lane models are generated using observed points from sensors, then lane path information is obtained, but uncertainty in the observed points propagates to uncertainty in the lane models
Solution Approach 1:
The patent segments the lane modeling problem into two independent components: a mean lane model representing the central path and an uncertainty model representing the confidence levels. This segmentation allows each component to be modeled separately using dedicated Recursive Least Squares filters, preventing uncertainty propagation from degrading the mean path accuracy while providing explicit uncertainty quantification for navigation decisions.
2Loss of information
If uncertainty values are tracked for each observation point, then uncertainty quantification is improved, but system complexity increases
Solution Approach 1:
The patent merges the tracking of mean lane positions and uncertainty values into a unified Recursive Least Squares framework. By combining these functions into a single probabilistic model, the system maintains complete information (both mean and variance) without requiring separate complex processing pipelines, thus reducing overall system complexity while preserving full uncertainty information.
3Measurement precision
If multiple sensors are used to obtain observation points, then measurement accuracy is improved, but data processing complexity increases
Solution Approach 1:
The patent implements a universal Recursive Least Squares filtering framework that can process observation data from multiple sensor types (camera, lidar, radar) through a single unified algorithm. This multi-functional approach allows the system to integrate heterogeneous sensor data with different characteristics and uncertainty levels without requiring separate processing pipelines for each sensor type, thereby reducing processing complexity while maintaining high measurement precision.
Data Source
AI summary
Systems and methods involve obtaining observation points of a lane line using one or more sensors of a vehicle. Each observation point indicates a location of a point on the lane line. A method includes obtaining uncertainty values, each uncertainty value corresponding with one of the observation points. A lane model is generated or updated using the observation points. The lane model indicates a path of the lane line. An uncertainty model is generated or updated using the uncertainty values corresponding with the observation points. The uncertainty model indicates uncertainty associated with each portion of the lane model.