Autonomous Driving Scenario Generation for Recognition Testing Gaps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing scenario-based testing methods for autonomous driving systems focus primarily on determination and control, neglecting the importance of recognition testing, especially in dynamic driving scenarios.
Innovation Solution
An autonomous driving test scenario generation method that includes generating tracked objects, determining and supplementing non-tracked objects, reflecting non-tracking factors on non-tracked object data, and identifying elements within the operational design domain to create comprehensive test scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If scenario-based testing focuses only on determination and control functions, then the testing process is simpler and more focused, but the recognition function testing is neglected and incomplete
Solution Approach 1:
The scenario generation process is segmented into distinct functional modules: tracked object generation unit, non-tracked object determination unit, non-tracking factor reflection unit, element identification unit, and scenario generation unit. Each module handles a specific aspect of scenario creation, allowing comprehensive testing coverage while maintaining manageable complexity through modular design.
Solution Approach 2:
The scenario generation system is designed to handle multiple testing functions simultaneously - it generates scenarios that test recognition functions (by incorporating non-tracked objects with various non-tracking factors), determination functions, and control functions. This multi-functional approach ensures complete testing coverage without requiring separate testing systems for each function.
2Reliability
If comprehensive test scenarios including recognition, determination, and control functions are generated, then the evaluation is more thorough, but the scenario generation process becomes more complex
Solution Approach 1:
The system divides comprehensive scenario generation into specialized sub-units: tracked object generation for standard scenarios, non-tracked object determination for recognition testing, non-tracking factor reflection for environmental variability, element identification for operational design domain mapping, and scenario generation for final synthesis. This segmentation enables thorough evaluation while keeping each component's complexity manageable.
Solution Approach 2:
The system performs preliminary actions by pre-processing tracked object data, determining non-tracked objects, and reflecting non-tracking factors before final scenario generation. This preparatory processing ensures that when comprehensive scenarios are generated, all necessary elements are already prepared, reducing the complexity of the final synthesis process.
3Adaptability or versatility
If data from autonomous driving vehicles is processed to generate test scenarios, then the scenarios are more realistic and relevant, but the data processing requirements increase
Solution Approach 1:
The system extracts specific relevant information from autonomous driving vehicle data through dedicated processing units. The tracked object generation unit extracts and processes tracked object data, while the non-tracked object determination unit identifies objects that were not tracked. This selective extraction creates realistic test scenarios without requiring processing of all raw data, balancing realism with processing complexity.
4Measurement precision
If non-tracked objects and non-tracking factors are incorporated into scenario generation, then recognition testing is improved, but the scenario generation process becomes more complex
Solution Approach 1:
The non-tracked object determination unit and non-tracking factor reflection unit perform preliminary identification and characterization of non-tracked objects before scenario generation. By pre-determining which objects were not tracked and what non-tracking factors are present, the system improves recognition testing accuracy while reducing the complexity of the final scenario generation process, as these elements are already prepared.
Data Source
AI summary
There are provided a system and a method for generating data-based autonomous driving test scenarios for testing and evaluation of autonomous driving system. An autonomous driving test scenario generation method according to an embodiment generates tracked objects by using tracked object data that an autonomous driving vehicle collects while traveling, determines some non-tracked objects among the tracked objects and reflects a non-tracking factor on non-tracked object data, and generates an autonomous driving test scenario by using elements within an operational design domain of an autonomous test scenario and the non-tracked object data reflecting the non-tracking factor. Accordingly, scenarios for testing not only determination, control functions but also a recognition function are generated based on data acquired by autonomous driving, so that more inclusive, detailed testing and evaluation can be provided for autonomous driving systems performing dynamic driving tasks.