Vehicle Risk Model Adaptation for ADS Corner Cases
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Automated Driving Systems (ADS) face challenges in determining risk models for complex traffic scenarios, particularly 'corner cases' and edge situations, which are not adequately addressed by existing methods, leading to safety and performance concerns.
Innovation Solution
A method for dynamically and adaptively determining risk models by obtaining sensor data from vehicles, estimating risk values based on risk-associated parameters, and updating baseline risk models to form an acceptable risk model for forming driving policies, which can be used to improve safety and performance of ADS systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If statistical brute force validation is used to prove safety, then safety can be demonstrated given sufficient driving miles, but the method is infeasible due to the enormous computational resources and time required
Solution Approach 1:
The patent segments the validation process by dividing the complex driving scenario into multiple dimensions (traffic rules compliance, safety risk, performance metrics). Instead of validating entire driving sequences through brute force, the system evaluates individual scenarios against segmented criteria, making the validation process computationally feasible while maintaining comprehensive safety assessment
Solution Approach 2:
The patent performs preliminary action by pre-defining safety thresholds, risk models, and evaluation criteria before actual validation. The system establishes acceptable risk models and safety boundaries in advance, allowing subsequent driving scenarios to be evaluated against these pre-established standards rather than requiring exhaustive re-validation of each scenario
2Reliability
If formal methods are used to prove ADS safety, then safety can be proven given certain assumptions about operational environment, but the method is limited by restrictive assumptions that do not reflect real-world corner cases
Solution Approach 1:
The patent applies dynamics by transitioning from static formal verification with fixed assumptions to a dynamic risk evaluation system. The system continuously updates risk models and safety assessments based on actual driving scenarios, allowing it to adapt to corner cases and unexpected situations while maintaining safety proofs through evolving risk thresholds and learned behavior patterns
Solution Approach 2:
The patent implements feedback mechanisms where actual driving data from corner cases and edge scenarios is fed back into the risk model refinement process. This feedback loop allows the system to learn from real-world deviations from assumed conditions and update safety models accordingly, bridging the gap between formal method assumptions and real-world variability
3Reliability
If a conservative risk model is used to ensure safety, then safety margins are increased, but the ADS performance and driving efficiency deteriorate due to excessive caution
Solution Approach 1:
The patent applies local quality by implementing spatially and situationally varying risk thresholds. Instead of applying a uniform conservative risk model across all scenarios, the system adjusts safety margins locally based on specific location characteristics, traffic conditions, and scenario types. This allows the ADS to maintain high safety standards in高风险 areas while operating more efficiently in low-risk situations
Solution Approach 2:
The patent utilizes parameter changes by dynamically adjusting risk model parameters based on learned patterns from training data. The system modifies safety thresholds, risk weights, and decision boundaries according to the specific driving context, allowing it to optimize the balance between safety margins and driving efficiency for different operational conditions rather than maintaining fixed conservative parameters
4Reliability
If an ADS operates with high safety standards in all situations, then safety is maximized, but the system becomes overly conservative and fails to match human driver performance in normal conditions
Solution Approach 1:
The patent makes the safety behavior dynamic by learning from human driver patterns in normal conditions. The system adapts its risk tolerance and safety responses to match human-like driving behavior in routine situations, while maintaining elevated safety standards through continuous monitoring and intervention when actual risk is detected, creating a more natural and less robotic driving style
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure relates to methods, systems, a vehicle and a computer-readable storage medium and a computer program product. The method comprises obtaining a baseline risk model generated based on one or more risk-associated parameters. The method further comprises obtaining sensor data from a sensor system of a vehicle. The method further comprises estimating one or more risk values associated with the one or more risk-associated parameters based on the obtained sensor data and generating, based on the one or more estimated risk values, an adopted risk model. Further, the method comprises provisioning the generated adopted risk model for determining an acceptable risk model for forming a driving policy of an ADS.