Federated Kalman Filter for Automotive Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automotive positioning systems face challenges in achieving stable and reliable high-precision positioning due to the limitations of individual sensors, such as laser radars, cameras, and GNSS, particularly in adverse weather conditions and complex environments like underpasses, beside tall buildings, or in tunnels.
Innovation Solution
A positioning method utilizing a Federated Kalman filter to acquire and weigh the credibility of multiple positioning subsystems, including combined navigation, laser point cloud, and camera visual positioning, by calculating and adjusting information distribution weight coefficients based on real-time reliability, enabling effective fusion of data from these subsystems to optimize positioning accuracy and robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple positioning subsystems are used for fusion, then positioning accuracy and reliability are improved, but system complexity increases
Solution Approach 1:
The patent divides the positioning system into multiple independent subsystems (laser radar subsystem, visual positioning subsystem, GNSS subsystem, inertial navigation subsystem), each with dedicated processing channels. This segmentation allows each subsystem to operate independently while contributing to the overall positioning function, improving reliability without overwhelming system complexity through modular architecture.
Solution Approach 2:
The patent combines multiple positioning subsystems into a unified federated Kalman filter framework that processes data from all subsystems simultaneously. The main filter integrates results from multiple sub-filters, each handling a specific positioning subsystem, thereby achieving improved positioning reliability through data fusion while managing complexity through structured integration.
2Measurement precision
If dynamic weight adjustment is implemented based on real-time credibility, then positioning accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent implements dynamic weight adjustment by calculating real-time credibility for each positioning subsystem based on current operating conditions and historical performance. The information distribution weight coefficients are continuously updated according to real-time credibility assessments, allowing the system to adapt to changing environmental conditions and maintain optimal positioning accuracy dynamically.
Solution Approach 2:
The patent employs a feedback mechanism where the main filter provides information distribution weight coefficients back to sub-filters based on global positioning data. This feedback loop allows the system to learn from past performance and adjust weights optimally, improving positioning accuracy while managing computational complexity through iterative refinement rather than exhaustive calculation.
3Reliability
If Federated Kalman filter fusion is performed, then robustness in adverse conditions is improved, but processing time increases
Solution Approach 1:
The patent segments the fusion process into parallel sub-filters, each processing data from a specific positioning subsystem independently. This parallel processing structure allows simultaneous computation of multiple positioning streams without sequential bottlenecks, reducing overall processing time while maintaining the robustness benefits of comprehensive data fusion through the main filter integration.
Data Source
AI summary
A positioning method includes: acquiring the credibility of each positioning subsystem in different states and generating a credibility data table; acquiring the real-time credibility from the corresponding credibility data table according to real-time positioning data of each positioning subsystem; calculating a first information distribution weight coefficient of each positioning subsystem involving a fusion operation of an filter according to the real-time credibility of each positioning subsystem; respectively feeding back, by a main filter, a second information distribution weight coefficient of each positioning subsystem involving the fusion operation to each sub-filter according to global data; determining a final information distribution weight coefficient of each positioning subsystem involving the fusion operation according to the first information distribution weight coefficient and the second information distribution weight coefficient; and performing, by the filter, the fusion operation according to the final information distribution weight coefficient of each positioning subsystem and outputting a final positioning result.