Kalman Filter Loss Scaling for Stable Autonomous Vehicle State Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face accuracy and precision issues due to sensor errors, which can impact navigation and safety, particularly under adverse conditions such as environmental interference or sensor malfunctions.
Innovation Solution
Incorporating a loss component into Kalman filters to dampen the effect of large measurement errors and ensuring symmetric, positive definite uncertainties through matrix factorization, thereby improving accuracy and preventing system destabilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensor data is used for autonomous vehicle navigation, then the vehicle can traverse environments autonomously, but sensor errors and measurement inaccuracies can cause large deviations in state estimation leading to unsafe operations
Solution Approach 1:
The patent introduces a loss component as an intermediary element between the sensor measurements and the Kalman filter state estimation. This loss component acts as a mediator that processes the raw sensor data first, transforming it in a way that reduces the impact of large measurement errors before the data enters the standard Kalman filter estimation process, thereby improving safety without requiring more precise sensors
Solution Approach 2:
The patent converts the harmful effect of large sensor measurement errors into a beneficial outcome by using a loss component that specifically targets and dampens the influence of outliers. Instead of treating measurement errors as purely negative, the system uses the loss component to identify and reduce the weight of erroneous measurements, transforming what would be harmful deviations into opportunities for more robust state estimation
2Ease of operation
If standard Kalman filter is used to estimate vehicle state, then computation is straightforward, but large measurement errors can cause unsolvable states and system destabilization
Solution Approach 1:
The patent applies preliminary action by introducing the loss component before the standard Kalman filter estimation process. This preprocessing step modifies the measurement data in advance to prevent potential instability issues, ensuring that the subsequent Kalman filter operations remain computationally simple while operating on more reliable, error-dampened data
3Loss of information
If measurement errors are fully trusted in Kalman filter update, then estimation uses all available information, but large errors can dominate and destabilize the system
Solution Approach 1:
The patent applies local quality by making the treatment of measurement information non-uniform. The loss component selectively processes different measurements based on their error characteristics, applying different levels of damping to different data points. This allows the system to fully utilize reliable measurements while locally reducing the influence of erroneous ones, maintaining both information utilization and stability
Data Source
AI summary
The techniques discussed herein include modifying a Kalman filter to additionally include a loss component that dampens the effect measurements with large errors (or measurements indicating states that are rather different than the predicted state) have on the Kalman filter and, in particular, the updated uncertainty and/or updated prediction. In some examples, the techniques include scaling a Kalman gain based at least in part on a loss function that is based on the innovation determined by the Kalman filter. The techniques additionally or alternatively include a reformulation of a Kalman filter that ensures that the uncertainties determined by the Kalman filter remain symmetric and positive definite.