Neural Network Approximation of Critical Autonomous Driving Test Cases
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for verifying and validating autonomous driving systems require extensive testing, including actual vehicle operation, which is time-consuming and costly, especially for determining critical test cases in scenario-based testing.
Innovation Solution
A computer-implemented method using an artificial neural network to approximate critical test results for autonomous driving systems by defining a state space of driving situation parameters and iteratively approximating function values until a predetermined threshold is met, allowing for efficient identification of critical test cases without extensive simulation or real-world testing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If extensive testing including actual vehicle operation is performed to verify and validate autonomous driving systems, then the reliability and completeness of test results is improved, but the time consumption and cost increase significantly
Solution Approach 1:
The patent creates virtual copies of driving scenarios, test environments, and vehicle behaviors through simulation. Instead of physically testing every possible situation, the system generates virtual test cases that replicate real-world conditions, allowing comprehensive validation without the time and cost of actual vehicle operation for each test case
Solution Approach 2:
The system performs preliminary analysis to identify critical test cases that are most likely to reveal system failures or edge cases. By pre-selecting these high-value test scenarios before execution, the system maximizes validation effectiveness while minimizing the total number of tests required, thereby reducing overall testing time
2Reliability
If a large number of potentially possible driving situations are tested to ensure comprehensive validation, then the coverage and reliability of the validation is improved, but the effort and cost increase significantly
Solution Approach 1:
The system varies key parameters such as environmental conditions, traffic participant behaviors, and vehicle states to generate diverse test scenarios. By systematically changing these parameters within defined ranges, the system achieves comprehensive coverage of the state space without manually designing every possible test case, thereby improving efficiency
Solution Approach 2:
The patent divides the overall validation process into segments: critical case identification, virtual test execution, and result analysis. This segmentation allows each phase to be optimized independently, with automated tools handling case generation and execution, thereby improving overall testing efficiency while maintaining comprehensive coverage
3Measurement precision
If actual vehicle operation is required to determine test control instructions and environmental data, then the realism and accuracy of test results is improved, but the complexity and resource requirements of the testing process increase
Solution Approach 1:
The patent introduces virtual test environments and simulation models as intermediaries between the vehicle system under test and the validation process. These virtual components generate and process test data without requiring actual vehicle operation for each test case, thereby maintaining data accuracy while reducing the complexity of coordinating physical testing resources
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The invention relates to a computer-implemented method for approximating a subset of test results from a virtual test of a device for at least partial autonomous driving of a motor vehicle. The invention further relates to a test unit (1) for approximating a subset of test results from a virtual test of a device for at least partial autonomous driving of a motor vehicle. The invention also relates to a computer program and a computer-readable data carrier.