Abstract:Advancements in Autonomous Driving Systems (ADS) have brought significant benefits, but also raised concerns regarding their safety. Virtual tests are common practices to ensure the safety of ADS because they are more efficient and safer compared to field operational tests. However, capturing the complex dynamics of real-world driving environments and effectively generating risk scenarios for testing is challenging. In this paper, we propose a novel paradigm shift towards utilizing Causal Bayesian Networks (CBN) for scenario generation in ADS. The CBN is built and validated using Maryland accident data, providing a deeper insight into the myriad factors influencing autonomous driving behaviors. Based on the constructed CBN, we propose an algorithm that significantly enhances the process of risk scenario generation, leading to more effective and safer ADS. An end-to-end testing framework for ADS is established utilizing the CARLA simulator. Through experiments, we successfully generated 89 high-risk scenarios from 5 seed scenarios, outperforming baseline methods in terms of time and iterations required.

What problem does this paper attempt to address?

### Problems Addressed by the Paper The paper aims to address the challenges of generating realistic risk scenarios during the testing of Autonomous Driving Systems (ADS). Specifically, the paper focuses on the following points: 1. **Capturing Complex Dynamic Environments**: The driving environment in the real world is highly complex and variable. Existing testing methods struggle to effectively capture this complexity, resulting in less realistic test scenarios. 2. **Efficient Generation of Risk Scenarios**: Current methods often require multiple iterations to generate risk scenarios, which is time-consuming. Additionally, the number of generated scenarios is limited, failing to comprehensively cover various potential risk situations. 3. **Subjectivity and Limitations**: Present methods for generating test scenarios mainly rely on specific rules or expert experience, introducing subjective assumptions and limiting the ability to cover a wide range of complex scenarios. 4. **Single-Dimension Evaluation**: Existing methods for evaluating driving behavior typically assess from a single dimension, lacking a comprehensive understanding of the complexity of actual driving behavior. ### Solution To address the above issues, the paper proposes a new framework based on Causal Bayesian Networks (CBN) for generating risk scenarios for Autonomous Driving Systems. The specific innovations include: 1. **Constructing CBN Based on Real Accident Data**: By using data mining and causal inference algorithms, the framework automatically identifies key risk factors related to behavior patterns and environmental conditions, reducing subjective assumptions in constructing test scenarios. 2. **Defining CBN-Based Scenario Generation Strategies**: Compared to baseline methods, this approach can efficiently generate risk scenarios in fewer iterations. 3. **Introducing Comprehensive Driving Behavior Evaluation Methods**: The framework evaluates driving behavior from multiple dimensions such as safety, comfort, and efficiency, providing a more comprehensive understanding and assessment. ### Experimental Validation The paper validates the effectiveness and efficiency of the proposed method through experiments. Using the CARLA simulator, an end-to-end testing framework was established, successfully generating 89 high-risk scenarios from 5 initial scenarios, significantly outperforming baseline methods in terms of time and number of iterations. ### Conclusion By constructing and applying Causal Bayesian Networks, the paper addresses the issues present in existing methods for generating realistic risk scenarios, improving the efficiency and effectiveness of testing Autonomous Driving Systems. This contributes to ensuring the safety and reliability of Autonomous Driving Systems.

Risk Scenario Generation for Autonomous Driving Systems based on Causal Bayesian Networks

Adversarial Safety-Critical Scenario Generation using Naturalistic Human Driving Priors

A Multimodal Data-Driven Approach for Driving Risk Assessment

Realistic Safety-critical Scenarios Search for Autonomous Driving System via Behavior Tree

Critical Scenario Generation for Realistic Testing of Autonomous Driving System Via Multi-vehicle Cooperative Control

Bridging Data-Driven and Knowledge-Driven Approaches for Safety-Critical Scenario Generation in Automated Vehicle Validation

Generating Edge Cases for Testing Autonomous Vehicles Using Real-World Data

A Survey on Safety-Critical Driving Scenario Generation -- A Methodological Perspective

A Survey on Safety-Critical Driving Scenario Generation—A Methodological Perspective

CausalAF: Causal Autoregressive Flow for Safety-Critical Driving Scenario Generation

Test scenario generation method for autonomous vehicles based on combinatorial testing and Bayesian network

Generating Critical Scenarios for Testing Automated Driving Systems

Adversarial Generation of Safety-Critical Lane-Change Scenarios for Autonomous Vehicles

A Survey on Scenario-Based Testing for Automated Driving Systems in High-Fidelity Simulation

Integrated Scenario-based Analysis: A data-driven approach to support automated driving systems development and safety evaluation

Risk Assessment of Roundabout Scenarios in Virtual Testing Based on an Improved Driving Safety Field

LMM-enhanced Safety-Critical Scenario Generation for Autonomous Driving System Testing From Non-Accident Traffic Videos

A Review on Scenario Generation for Testing Autonomous Vehicles

Interactive Critical Scenario Generation for Autonomous Vehicles Testing Based on In-depth Crash Data Using Reinforcement Learning

Online Adaptive Generation of Critical Boundary Scenarios for Evaluation of Autonomous Vehicles

A Hazardous Lane-Changing Scenario Search Approach Based on Adaptive Random Testing for Automated Vehicles