Autonomous Vehicle Scenario Generation for Accident Avoidability Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for testing autonomous vehicles (AVs) are inefficient due to the complexity of their software, which includes deep neural networks, and traditional testing methods like formal verification and code coverage are not readily applicable, while existing simulation methods often fail to provide comprehensive scenarios for accident avoidance analysis.
Innovation Solution
A comprehensive technique for characterizing and generating driving scenarios based on multivariate metrics such as safe paths, effort, and time required for action, allowing for the creation of scenarios that assess the difficulty and risk of accident avoidance, independent of driver models and applicable to both simulators and real-world scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human experts specify driving scenarios for testing, then scenario quality can be ensured, but the process is time-consuming and difficult to scale
Solution Approach 1:
The system enables automated scenario generation where the computer automatically creates and characterizes driving scenarios using algorithms that evaluate safe paths, effort, and time metrics without requiring human expert intervention for each scenario creation
Solution Approach 2:
The patent replaces the manual mechanical process of human expert scenario specification with an automated computational system that uses algorithms to generate and evaluate scenarios based on defined metrics for safety, difficulty, and risk
2Extent of automation
If reinforcement learning or optimization methods are used to generate scenarios, then scenario generation can be automated, but the scenarios often do not provide information on whether accidents are avoidable
Solution Approach 1:
The system incorporates feedback mechanisms where scenarios are evaluated against multiple metrics including safe path count, narrowness, effort, and time required. This feedback loop ensures that generated scenarios provide meaningful information about accident avoidability and are prioritized accordingly for training
Solution Approach 2:
The patent replaces conventional automated generation methods (reinforcement learning, optimization) with a novel metric-based characterization system that explicitly evaluates and prioritizes scenarios based on their informational value regarding accident avoidability
3Reliability
If comprehensive safety testing is performed to ensure AV safety, then testing thoroughness improves, but the complexity of AV software including deep neural networks makes testing challenging
Solution Approach 1:
The patent segments the complex testing problem into manageable components by characterizing scenarios based on distinct metrics (safe paths, narrowness, effort, time) and prioritizing them systematically, making the overall testing process more tractable despite software complexity
Solution Approach 2:
The system changes the parameters of scenario evaluation from traditional binary safe/unsafe classification to a multi-dimensional metric space including safe path count, narrowness, effort, and time required, enabling more nuanced safety assessment of complex AV software
Data Source
AI summary
Techniques to generate driving scenarios for autonomous vehicles characterize a path in a driving scenario according to metrics such as narrowness and effort. Nodes of the path are assigned a time for action to avoid collision from the node. The generated scenarios may be simulated in a computer.