Autonomous Driving Sensor Fusion With Selective State Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-sensor fusion algorithms in autonomous driving fail to effectively suppress certain state dimensions without affecting other states, particularly when using Kalman filters, as they either ignore the impact of non-updated dimensions on covariance or force suppression without consideration.
Innovation Solution
Implementing a multi-sensor fusion method where state transition and Jacobian matrices for suppressed state quantities are set to zero matrices during Kalman filtering, while maintaining the covariance of non-suppressed quantities by setting Kalman gain matrices to zero, allowing flexible suppression without affecting the main state update.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Schmitt trigger method is used to suppress state dimensions, then state suppression is achieved, but the impact of non-updated dimensions on covariance of other dimensions is not considered, resulting in ineffective suppression
Solution Approach 1:
The patent changes the parameter values in the state transition matrix by setting specific elements to zero to prevent updates of certain state dimensions while maintaining the mathematical relationships for covariance propagation. This allows proper consideration of how non-updated dimensions affect other dimensions without completely isolating them.
2Reliability
If state and covariance are forced not to be updated, then state suppression is achieved, but the impact of covariance is not considered
Solution Approach 1:
The patent selectively changes parameters in the state transition matrix by setting specific elements to zero, allowing the covariance to be updated through proper mathematical propagation while preventing actual state updates for suppressed dimensions. This resolves the contradiction by maintaining covariance information flow without updating suppressed states.
3Measurement precision
If Kalman filter updates all state dimensions, then comprehensive state estimation is achieved, but state suppression for specific dimensions cannot be implemented
Solution Approach 1:
The patent applies local quality by treating different state dimensions differently through the state transition matrix. Specific elements are set to zero to suppress certain dimensions while maintaining normal update behavior for other dimensions. This allows selective suppression without affecting the overall estimation accuracy of non-suppressed dimensions.
4Ease of operation
If state suppression is implemented without considering covariance propagation, then computational simplicity is achieved, but suppression effectiveness is compromised
Solution Approach 1:
The patent implements suppression by simply changing parameter values (setting matrix elements to zero) in the state transition matrix, which is computationally simple. However, this simple approach naturally and correctly propagates through to the covariance calculations, maintaining suppression effectiveness without complex additional computations.
Data Source
Figure 1
AI summary
The disclosure relates to the technical field of autonomous driving, and specifically provides a multi-sensor fusion method based on autonomous driving, an apparatus, a medium, and a vehicle, which are intended to solve the problem of how to effectively suppress corresponding state quantities when some state dimensions need to be disabled during multi-sensor fusion, without affecting other states. In the method, when there is a first target state quantity that requires state suppression, in a propagation phase of Kalman filtering, a state transition matrix for the first target state quantity's self-dimension is set as an identity matrix, and state transition matrices for dimensions related to both the first target state quantity and a second target state quantity are set as a zero matrix; and in an update phase, a Jacobian matrix of the first target state quantity is set as a zero matrix, and a Kalman gain matrix of the first target state quantity is set as a zero matrix, thereby avoiding an impact of suppression of the first target state quantity on covariance of the second target state quantity.