Modeling Driving Scenarios in Road Traffic using Fleet Data from Production Vehicles

Centrala begrepp

This research project aims to develop a method for modeling the probabilities of occurrence of concrete driving scenarios across the statistical population of road traffic using data collected from production vehicles.

Sammanfattning

The paper discusses the need for statistical information and data from road traffic to meet the requirements of the ISO 21448 standard on the "Safety of the Intended Functionality" (SOTIF) for automated driving systems. It outlines the current state of research on modeling driving scenarios and proposes a method for abstracting and representing driving data collected from production vehicles.
The key steps of the proposed method are:

Qualitative abstraction of driving data recorded by vehicle sensors to identify logical driving actions and scenarios.
Quantitative abstraction by parameterizing the logical elements to enable accurate reconstruction of the driving dynamics.
Deriving the distributions and correlations of the scenario parameters to represent the probabilities of occurrence across the statistical population of road traffic.

The paper discusses the technical limitations and challenges in implementing this approach, such as sensor capabilities, data storage, and wireless transmission. It also highlights the need to extend existing knowledge representation techniques like ontologies to include the probabilistic aspects of driving scenarios.

Statistik

"The ultimate goal of the SOTIF activities is to evaluate the potentially hazardous behaviour in hazardous scenarios and to provide an argument that the residual risk caused by these scenarios is sufficiently low."
"The probability of occurrence of these hazardous scenarios must be known."
"The whole set of V&V activities shall cover the possible scenario space representatively."

Citat

"The absence of unreasonable risk resulting from hazardous behaviours related to [such] functional insufficiencies is defined as the safety of the intended functionality (SOTIF) by ISO 21448."
"The size of the areas represents the number of scenarios, not the risk posed by those scenarios."
"The verification and validation strategy as well as validation targets shall be defined and shall consider the sufficient coverage of the relevant scenario space."

Viktiga insikter från

Statistical Modelling of Driving Scenarios in Road Traffic using Fleet Data of Production Vehicles

by Chri... på arxiv.org 04-10-2024

https://arxiv.org/pdf/2404.06288.pdf

Djupare frågor

How can the proposed method be extended to incorporate other relevant data sources beyond production vehicle fleet data, such as traffic simulation, infrastructure sensors, and accident reports, to improve the representativeness of the modeled driving scenarios?

To enhance the representativeness of the modeled driving scenarios, the proposed method can be extended by integrating various additional data sources.

Traffic Simulation Data: By incorporating data from traffic simulation models, the system can account for a wider range of scenarios that may not be captured solely by production vehicle data. This would provide a more comprehensive understanding of complex traffic interactions and scenarios that are not frequently encountered in real-world driving.

Infrastructure Sensors: Leveraging data from infrastructure sensors such as traffic cameras, road sensors, and connected infrastructure systems can offer valuable insights into the environment surrounding the vehicles. This data can help in modeling scenarios influenced by road conditions, traffic signals, and pedestrian movements, contributing to a more holistic representation of driving scenarios.

Accident Reports: Analyzing historical accident reports and incident data can provide crucial information on high-risk scenarios and contributing factors to accidents. By integrating this data, the system can prioritize modeling scenarios that have a higher probability of leading to hazardous situations, thereby improving the safety assessment process.

Naturalistic Driving Studies: Incorporating data from naturalistic driving studies, which involve the continuous monitoring of real-world driving behavior, can offer valuable insights into driver actions, reactions, and the prevalence of specific driving scenarios. This data can enrich the modeling process by providing a more nuanced understanding of real-world driving dynamics.

By combining data from these diverse sources, the method can create a more robust and comprehensive representation of driving scenarios, encompassing a broader spectrum of potential risks and interactions on the road.

What are the potential biases and limitations in using production vehicle data, and how can they be addressed to ensure the validity of the statistical models?

While production vehicle data is valuable for modeling driving scenarios, it comes with certain biases and limitations that need to be addressed to ensure the validity of the statistical models:

Sampling Bias: Production vehicle data may not fully represent the entire spectrum of driving scenarios, as it is limited to the behavior of specific vehicle models in certain conditions. To mitigate sampling bias, data collection efforts should aim to capture a diverse range of driving scenarios across different locations, traffic conditions, and driving styles.

Sensor Limitations: The accuracy and coverage of sensors in production vehicles may vary, leading to gaps or inaccuracies in the recorded data. Calibration issues, sensor failures, or limited sensor range can introduce biases in the data. Regular sensor maintenance, calibration checks, and data validation processes are essential to address these limitations.

Privacy Concerns: Production vehicle data often contains sensitive information about drivers and passengers. Ensuring data anonymization and compliance with data privacy regulations is crucial to protect individual privacy rights while using the data for statistical modeling.

Data Quality: Inaccuracies, missing data, or inconsistencies in production vehicle data can impact the reliability of statistical models. Implementing data quality checks, validation procedures, and data cleaning techniques can help improve the overall quality and integrity of the dataset.

By addressing these biases and limitations through robust data collection practices, quality assurance measures, and privacy safeguards, the validity and reliability of the statistical models based on production vehicle data can be enhanced.

How can the probabilistic driving scenario models be integrated with other components of the safety assurance process, such as hazard analysis and risk assessment, to provide a comprehensive argument for the safety of the intended functionality?

Integrating probabilistic driving scenario models with hazard analysis and risk assessment processes is essential to establish a comprehensive safety assurance framework for the intended functionality of automated driving systems. Here's how this integration can be achieved:

Scenario-Based Hazard Analysis: Utilize the probabilistic driving scenario models to identify potential hazards and critical scenarios that could lead to safety risks. By mapping these scenarios to specific hazards and assessing their severity and likelihood, the hazard analysis process can be enriched with real-world driving data, enhancing the identification of safety-critical situations.

Risk Assessment Incorporating Scenario Probabilities: Integrate the probabilities derived from the driving scenario models into the risk assessment framework. By quantifying the likelihood of different scenarios and their associated risks, the risk assessment process can prioritize mitigation strategies for high-risk scenarios and allocate resources effectively to address safety concerns.

Validation and Verification: Use the probabilistic driving scenario models to validate the safety of the intended functionality through scenario-based testing. By simulating a wide range of driving scenarios based on their probabilities, the system can verify the robustness of the automated driving functions and ensure that they perform safely under various conditions.

Continuous Monitoring and Improvement: Establish a feedback loop between the probabilistic driving scenario models and the safety assurance processes to continuously monitor performance, identify new hazards, and refine risk mitigation strategies. Regular updates to the scenario models based on real-world data can enhance the safety assurance framework over time.

By integrating probabilistic driving scenario models with hazard analysis, risk assessment, validation, and continuous improvement processes, a comprehensive argument for the safety of the intended functionality can be developed, ensuring that automated driving systems meet the required safety standards and perform reliably in diverse driving scenarios.