Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FREA: Feasibility-Guided Generation of Safety-Critical Scenarios with Reasonable Adversariality

Keyu Chen, Yuheng Lei, Hao Cheng, Haoran Wu, Wenchao Sun, Sifa Zheng

TL;DR

FREA is a novel safety-critical scenarios generation method that incorporates the Largest Feasible Region of AV as guidance to ensure the reasonableness of the adversarial scenarios, yielding considerable near-miss events while ensuring AV's feasibility.

Abstract

Generating safety-critical scenarios, which are essential yet difficult to collect at scale, offers an effective method to evaluate the robustness of autonomous vehicles (AVs). Existing methods focus on optimizing adversariality while preserving the naturalness of scenarios, aiming to achieve a balance through data-driven approaches. However, without an appropriate upper bound for adversariality, the scenarios might exhibit excessive adversariality, potentially leading to unavoidable collisions. In this paper, we introduce FREA, a novel safety-critical scenarios generation method that incorporates the Largest Feasible Region (LFR) of AV as guidance to ensure the reasonableness of the adversarial scenarios. Concretely, FREA initially pre-calculates the LFR of AV from offline datasets. Subsequently, it learns a reasonable adversarial policy that controls the scene's critical background vehicles (CBVs) to generate adversarial yet AV-feasible scenarios by maximizing a novel feasibility-dependent adversarial objective function. Extensive experiments illustrate that FREA can effectively generate safety-critical scenarios, yielding considerable near-miss events while ensuring AV's feasibility. Generalization analysis also confirms the robustness of FREA in AV testing across various surrogate AV methods and traffic environments.

FREA: Feasibility-Guided Generation of Safety-Critical Scenarios with Reasonable Adversariality

TL;DR

FREA is a novel safety-critical scenarios generation method that incorporates the Largest Feasible Region of AV as guidance to ensure the reasonableness of the adversarial scenarios, yielding considerable near-miss events while ensuring AV's feasibility.

Abstract

Generating safety-critical scenarios, which are essential yet difficult to collect at scale, offers an effective method to evaluate the robustness of autonomous vehicles (AVs). Existing methods focus on optimizing adversariality while preserving the naturalness of scenarios, aiming to achieve a balance through data-driven approaches. However, without an appropriate upper bound for adversariality, the scenarios might exhibit excessive adversariality, potentially leading to unavoidable collisions. In this paper, we introduce FREA, a novel safety-critical scenarios generation method that incorporates the Largest Feasible Region (LFR) of AV as guidance to ensure the reasonableness of the adversarial scenarios. Concretely, FREA initially pre-calculates the LFR of AV from offline datasets. Subsequently, it learns a reasonable adversarial policy that controls the scene's critical background vehicles (CBVs) to generate adversarial yet AV-feasible scenarios by maximizing a novel feasibility-dependent adversarial objective function. Extensive experiments illustrate that FREA can effectively generate safety-critical scenarios, yielding considerable near-miss events while ensuring AV's feasibility. Generalization analysis also confirms the robustness of FREA in AV testing across various surrogate AV methods and traffic environments.
Paper Structure (32 sections, 1 theorem, 15 equations, 11 figures, 6 tables, 1 algorithm)

This paper contains 32 sections, 1 theorem, 15 equations, 11 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Largest Feasible Region of AV
  5. Feasibility-Dependent Adversarial Objective of CBV
  6. Experiments
  7. Experiment Setup
  8. Largest Feasible Region Learning of AV
  9. Quantitative Analysis of Generated Scenarios
  10. Generalization Analysis of AV Testing.
  11. Analysis of Representative Scenarios
  12. Conclusion
  13. Theoretical Analysis
  14. Feasibility-depend Advantage Function
  15. Proof of Lemma \ref{['lemma:fea_Q']}
  16. ...and 17 more sections

Key Result

Lemma 1

As the BVs follow deterministic policy, the optimal feasible action-value function of AV can be achieved by AV's current state and next state (see ap:proof of lemma fea_Q for proof).

Figures (11)

  • Figure 1: Illustration of adversarial yet AV-feasible scenarios in a two-lane traffic setting. The CBV employs three distinct policies, resulting in different scenarios: (a) Conservative scenario, where the policy is less adversarial; (b) Excessive adversarial scenario, resulting in an unavoidable collision; and (c) Ideal adversarial scenario, effectively balancing adversarial and AV's feasibility.
  • Figure 2: Visualization of well-trained LFR in various scenarios.
  • Figure 3: Evaluating near-miss events across different CBV methods with Expert chitta2022transfuser as AV
  • Figure 4: Evaluating collision severity across different CBV methods with Expert chitta2022transfuser as AV
  • Figure 5: Representative scenarios: Red: AV (Expert chitta2022transfuser). Blue: BV. Purple: CBV (FREA).
  • ...and 6 more figures

Theorems & Definitions (3)

  • Definition 1: Optimal feasible value function
  • Definition 2: Largest Feasible Region (LFR)
  • Lemma 1