Multi-Lane Vehicle State Estimation Using Filter Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the state of a vehicle on a road, especially in multi-lane scenarios, face challenges in accuracy and reliability due to limitations in satellite-based positioning systems and measurement noise from various sensors, making it difficult to predict a robust vehicle pose estimation.
Innovation Solution
A system and method that initialize multiple filters per lane using map data, positioning data, and sensor data, and employ a trained machine learning algorithm to select the most accurate filter based on confidence probability values and multi-objective optimized coefficients, enhancing the accuracy and adaptability of vehicle state estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If satellite-based positioning systems (GNSS) are used for vehicle positioning, then positioning coverage is provided, but positioning accuracy is insufficient for autonomous applications
Solution Approach 1:
The patent combines multiple positioning systems (GNSS, inertial sensors, wheel encoders, camera-based vision systems) into an integrated pose determination system. This fusion of multiple independent positioning sources compensates for individual system limitations and achieves the required accuracy for autonomous driving applications.
Solution Approach 2:
The system uses a composite approach by integrating heterogeneous sensing technologies (satellite signals, inertial measurement units, optical cameras, encoder data) into a unified positioning solution, similar to how composite materials combine different substances to achieve superior properties.
2Reliability
If multiple filters are initialized per lane to improve state estimation accuracy, then measurement reliability is improved, but computational complexity increases
Solution Approach 1:
The system segments the road into multiple lanes and initializes separate filters for each lane. This segmentation allows the system to track and evaluate multiple hypotheses simultaneously, improving reliability by selecting the most probable lane state while managing computational load through structured organization.
Solution Approach 2:
The filter initialization and selection process is dynamic, adapting to changing road conditions and vehicle states. The system continuously evaluates multiple filter hypotheses and selects the most appropriate one based on current sensor data and contextual information.
3Measurement precision
If machine learning algorithms are used to select filters, then state estimation accuracy is improved, but processing time increases
Solution Approach 1:
The machine learning model is trained offline in advance on large datasets, so that during actual operation, the system only needs to infer from pre-processed sensor data. This preliminary training phase separates the computationally intensive learning process from the real-time decision-making process.
Solution Approach 2:
The patent replaces traditional rule-based or deterministic filter selection methods with machine learning-based probabilistic classification. This substitution enables more accurate state estimation by leveraging patterns learned from data, while the system is designed to manage the computational requirements through efficient implementation.
Data Source
Figure 1
Figure 2
Figure 3a
AI summary
The present disclosure relates to a method, system, a vehicle and a computer-readable storage medium for determining a state of a vehicle on a road portion having two or more lanes. The vehicle comprises an Automated Driving System (ADS) feature. The method comprises obtaining map data associated with the road portion and obtaining positioning data indicating a pose of the vehicle on the road and obtaining sensor data from a sensor system of the vehicle. Further, the method comprises initializing a plurality of filters for the road portion wherein one filter is initialized per lane of the road portion. The method further comprises associating one or more sensor data point(s) in the obtained sensor data to a corresponding map-element of the obtained map data and determining one or more normalized similarity score(s) between the associated obtained map data and the obtained sensor data. Even further the method comprises determining one or more multivariate time-series data based on the determined one or more normalized similarity score(s), wherein each multivariate time-series data is attributed to a corresponding initialized filter among the plurality of initialized filters. Further the method comprises providing the one or more multivariate time-series data as input to a trained machine-learning algorithm and selecting one of the initialized filters by means of the machine learning algorithm.