Autonomous Driving Risk Estimation Using Manual Driving Benchmarks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of autonomous driving situations poses challenges for artificial intelligence software, leading to potential accidents and liability issues, with extensive testing required to ensure adequate capability, which is time-consuming and prone to regression.
Innovation Solution
A method that determines autonomous-driving quantities from sensor data, compares them to manual-driving quantities, and adjusts a risk management model to assess and improve autonomous vehicle safety and insurance premiums, using a database of manual-driving data to simulate and predict accident rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If extensive testing is conducted to ensure adequate AI software capability, then reliability of autonomous driving is improved, but loss of time and productivity deteriorate
Solution Approach 1:
The patent applies preliminary action by pre-collecting and storing manual driving data from multiple drivers in a database before autonomous testing. This pre-prepared data serves as a reference framework that enables faster evaluation of autonomous driving performance, reducing the time needed for extensive real-world testing while maintaining reliability assessment accuracy.
Solution Approach 2:
The patent uses copying by creating virtual representations of manual driving behaviors through recorded data. Instead of physically testing every possible scenario with human drivers, the system copies and analyzes manual driving patterns from the database to evaluate and compare autonomous driving performance, significantly reducing testing time while preserving reliability assessment.
2Reliability
If extensive testing is conducted to ensure adequate AI software capability, then reliability of autonomous driving is improved, but productivity deteriorates
Solution Approach 1:
The patent implements feedback by systematically comparing autonomous driving quantities against manual driving quantities from the database. This comparison provides continuous feedback on AI software performance across multiple dimensions, enabling targeted improvements and reducing the need for exhaustive testing cycles, thereby improving software development productivity while maintaining reliability.
Solution Approach 2:
The patent applies preliminary action by pre-collecting and storing manual driving data from multiple drivers in a database before autonomous testing. This pre-prepared data serves as a reference framework that enables faster evaluation of autonomous driving performance, reducing the time needed for extensive real-world testing while maintaining reliability assessment accuracy.
3Reliability
If AI software is enhanced to handle complex driving situations, then reliability is improved, but device complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the evaluation of autonomous driving into distinct quantitative dimensions (e.g., acceleration patterns, braking behavior, steering responses). This segmented approach allows the complex AI software to be assessed and improved in manageable components, reducing overall system complexity while enhancing reliability through targeted capability development.
Data Source
AI summary
The present invention relates to a method for estimating an accident risk of an autonomous driving unit. The method produces helpful results with less autonomous driving cycles. An autonomous-driving quantity, quantifying an autonomous-driving quality of the driving of the driving unit, is determined from driving values that have been determined from monitoring at least one driving parameter of the driving unit during autonomous driving. The autonomous-driving quantity is associated with a plurality of manual-driving quantities that have been determined from the same driving parameter during manual driving periods of different driving units. An autonomous-driving accident rate value is determined from accident rate values associated with those manual-driving quantities.