Autonomous Driving Safety Evaluation with Segmented Test Spaces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current evaluation techniques for autonomous driving vehicles struggle to quantify safety due to the vast number of interactions with peripheral vehicles, leading to a combinatorial explosion and an impractical, infinite test space.
Innovation Solution
The method involves dividing the test space into an intended test space, where safety can be quantified, and an unintended test space, where characteristics are not quantifiable. Feedback from the unintended test space is applied to the intended test space using a combined simulation of peripheral and autonomous vehicles to evaluate safety.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional evaluation techniques are used to assess autonomous vehicle safety, then comprehensive coverage of interaction scenarios is attempted, but the test space becomes infinite and unmanageable due to combinatorial explosion
Solution Approach 1:
The patent segments the test space into two distinct parts: an intended test space containing quantifiable characterizations and an unintended test space containing non-quantifiable characterizations. This segmentation allows the evaluation system to manage the infinite test space by handling only the finite, quantifiable portion directly while representing the rest through feedback mechanisms, thereby resolving the contradiction between comprehensive safety evaluation and manageable test space complexity.
2Reliability
If test cases are generated based on interdependence between autonomous vehicle and peripheral vehicles, then safety coverage is improved, but combinatorial explosion occurs making the test space infinite
Solution Approach 1:
The patent introduces a feedback mechanism where the unintended test space (containing non-quantifiable characterizations) provides feedback to the intended test space (containing quantifiable characterizations). This feedback loop allows the system to account for the infinite combinations of test cases without explicitly generating them all, thereby improving safety evaluation coverage while avoiding combinatorial explosion by working with a finite representation.
Data Source
AI summary
A computer implemented method for evaluating autonomous vehicle safety that includes defining criteria for safety of autonomous vehicles in a test space, and dividing the test space into an intended test space and a un-intended test space for the criteria for safety of autonomous vehicles. The intended test space includes characterizations for the autonomous vehicle that can be quantified, and the un-intended test space includes characterizations that are not quantifiable. The method further includes measuring the safety of the autonomous vehicles in the intended test space. The applying the un-intended test space is applied to the intended test space as feedback into the intended test space; and evaluating the intended test space including the feedback from the unintended test space using a combined simulation of peripheral vehicles and autonomous vehicles to provide the evaluation of autonomous vehicle safety.