Vehicle Sensor Fusion Validation Using Residual Error Probability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for releasing autonomous vehicle sensor systems require an impractically high number of test kilometers to ensure safety, as existing statistical models are not tailored to perception systems and cannot differentiate between errors in object detection, making it difficult to validate the safety of autonomous vehicles.
Innovation Solution
A method that calculates the probability of failure by considering deviation combinations of individual sensors, integrating residual risk analysis and using a fusion unit to process and validate sensor data, thereby reducing the required test kilometers for sensor system release.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current statistical models are used for validating sensor systems, then safety requirements are met through extensive testing, but the number of test kilometers required becomes impractically high (billions of kilometers)
Solution Approach 1:
The patent segments the sensor system validation into distinct components: individual sensor deviations are analyzed separately, then combined through a fusion unit to assess overall system behavior. This segmentation allows for targeted analysis of specific sensor types and their error patterns, rather than requiring comprehensive testing of all possible system states
Solution Approach 2:
The patent applies preliminary action by using simulated deviation combinations to predict system behavior before actual deployment. By pre-calculating how various sensor errors might combine and affect detection outcomes, the system can validate safety without requiring exhaustive real-world testing
2Reliability
If existing approval concepts are applied to autonomous vehicles, then safety control is ensured through human driver fallback, but the complexity of approval procedures increases and legal frameworks are lacking
Solution Approach 1:
The patent implements feedback by continuously monitoring sensor deviations and comparing actual system performance against predicted error patterns. This feedback mechanism enables automated safety verification, reducing reliance on complex human-in-the-loop approval processes and providing objective data for regulatory compliance
Solution Approach 2:
The patent introduces an intermediary statistical framework that mediates between raw sensor data and safety certification requirements. This framework translates complex sensor behaviors into quantifiable risk metrics that can be evaluated against safety standards, simplifying the approval process
3Measurement precision
If comprehensive testing is performed to validate sensor system safety, then detection accuracy is verified, but the cost and time requirements become prohibitive for practical deployment
Solution Approach 1:
The patent applies parameter changes by systematically varying sensor deviation parameters in simulations to identify critical error combinations. By changing and analyzing specific deviation parameters rather than testing all possible scenarios, the system efficiently identifies which detection accuracy aspects require validation
Data Source
Figure 1~5
Figure 6~10
Figure 11~12
AI summary
The invention relates to a method for enabling a sensor system (10) for detecting objects in an environment of a vehicle (1), the method comprising the steps of: providing a common probability distribution for deviations (a) between output data from the sensor system (10) and reference data at the level of program sections for detecting objects of the sensor system (10), at the level of sensors (11) in the sensor system (10) and/or at the fusion level of the sensor system (10) (V1), sampling combinations of deviations and calculating probabilities of occurrence (p) for the sampled combinations of deviations by means of the common probability distribution (V2), applying the sampled combinations of deviations to the reference data, processing the reference data, to which said combinations of deviations have been applied, by means of a fusion unit (12) of the sensor system (10) and obtaining fusion results (V3), removing those probabilities of occurrence (p), the underlying combinations of deviations of which result in those fusion results which satisfy a predefined condition, from the common probability distribution and obtaining a residual probability distribution (V4), integrating the residual probability distribution and obtaining an absolute error probability (P) (V5), and enabling the sensor system (10) on the basis of the absolute error probability (P) (V6).