Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Conflicts to Collisions: A Two-Stage Collision Scenario-Testing Approach for Autonomous Driving Systems

Siyuan Chen, Fuyuan Zhang, Hua Qi, Lei Ma, Tomoyuki Tsuchiya, Michio Hayashi, Manabu Okada

TL;DR

This work proposes a two-stage scenario testing framework that first searches for conflicts and then mutates these conflict scenarios to induce actual collisions, showing that using conflicts as intermediate objectives broadens the search horizon and significantly improves the efficiency and effectiveness of ADS safety evaluation.

Abstract

Autonomous driving systems (ADS) are safety-critical and require rigorous testing before public deployment. Simulation-based scenario testing provides a safe and cost-effective alternative to extensive on-road trials, enabling efficient evaluation of ADS under diverse and high-risk conditions. However, existing approaches mainly evaluates the scenarios based on their proximity to collisions and focus on scenarios already close to collision, leaving many other hazardous situations unexplored. To bridge this, we introduce a collision-related concept of conflict as an intermediate search target and propose a two-stage scenario testing framework that first searches for conflicts and then mutates these conflict scenarios to induce actual collisions. Evaluated on Baidu Apollo, our approach reveals up to 12 distinct collision types in a single run, doubling the diversity discovered by state-of-the-art baselines while requiring fewer simulations thanks to conflict-targeted mutations. These results show that using conflicts as intermediate objectives broadens the search horizon and significantly improves the efficiency and effectiveness of ADS safety evaluation.

From Conflicts to Collisions: A Two-Stage Collision Scenario-Testing Approach for Autonomous Driving Systems

TL;DR

This work proposes a two-stage scenario testing framework that first searches for conflicts and then mutates these conflict scenarios to induce actual collisions, showing that using conflicts as intermediate objectives broadens the search horizon and significantly improves the efficiency and effectiveness of ADS safety evaluation.

Abstract

Autonomous driving systems (ADS) are safety-critical and require rigorous testing before public deployment. Simulation-based scenario testing provides a safe and cost-effective alternative to extensive on-road trials, enabling efficient evaluation of ADS under diverse and high-risk conditions. However, existing approaches mainly evaluates the scenarios based on their proximity to collisions and focus on scenarios already close to collision, leaving many other hazardous situations unexplored. To bridge this, we introduce a collision-related concept of conflict as an intermediate search target and propose a two-stage scenario testing framework that first searches for conflicts and then mutates these conflict scenarios to induce actual collisions. Evaluated on Baidu Apollo, our approach reveals up to 12 distinct collision types in a single run, doubling the diversity discovered by state-of-the-art baselines while requiring fewer simulations thanks to conflict-targeted mutations. These results show that using conflicts as intermediate objectives broadens the search horizon and significantly improves the efficiency and effectiveness of ADS safety evaluation.
Paper Structure (18 sections, 2 equations, 17 figures, 6 tables, 1 algorithm)

This paper contains 18 sections, 2 equations, 17 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Conflict-based Scenario Testing
  3. Space-sharing Conflicts
  4. Using Conflicts for Scenario Testing
  5. Technical Details
  6. Scenario Representation
  7. Overview
  8. Genetic Algorithm for Conflict Search
  9. Mutation-based Fuzzing for Collision Search
  10. Experiments
  11. Experiment Design
  12. Results for RQ1
  13. Result for RQ2
  14. Result for RQ3
  15. Result for RQ4
  16. ...and 3 more sections

Figures (17)

  • Figure 1: A visualization of collision types found by AV-Fuzzer over 100 consecutive search steps. Each vertical bar on the chart represents one search step. The gray background indicates a scenario where no collision occurred, while the colored bars highlight a successful collision scenario. Different colors correspond to the distinct collision types detailed in Appendix A.2. The chart shows that within this interval, the collisions found are of limited variety.
  • Figure 2: 5 types of conflicts.
  • Figure 3: An example of merging path conflict.
  • Figure 4: Representation of a scenario.
  • Figure 5: Overview of our approach.
  • ...and 12 more figures

Theorems & Definitions (1)

  • Definition 1