Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

3CSim: CARLA Corner Case Simulation for Control Assessment in Autonomous Driving

Matúš Čávojský, Eugen Šlapak, Matúš Dopiriak, Gabriel Bugár, Juraj Gazda

TL;DR

The paper tackles the challenge of evaluating autonomous driving systems under rare and cognitively demanding scenarios. It introduces 3CSim, a CARLA-based framework that enables deterministic, repeatable simulations for both assessing control performance and generating corner-case datasets. A three-part taxonomy—state anomalies, behavior anomalies, and evidence-based anomalies—maps 32 implemented corner-case scenarios with adjustable weather, timing, and traffic density. The work advances safety validation by providing a structured, extensible platform for systematic testing of AD control under non-standard conditions, with future work aimed at expanding scenarios and developing specialized evaluation metrics.

Abstract

We present the CARLA corner case simulation (3CSim) for evaluating autonomous driving (AD) systems within the CARLA simulator. This framework is designed to address the limitations of traditional AD model training by focusing on non-standard, rare, and cognitively challenging scenarios. These corner cases are crucial for ensuring vehicle safety and reliability, as they test advanced control capabilities under unusual conditions. Our approach introduces a taxonomy of corner cases categorized into state anomalies, behavior anomalies, and evidence-based anomalies. We implement 32 unique corner cases with adjustable parameters, including 9 predefined weather conditions, timing, and traffic density. The framework enables repeatable and modifiable scenario evaluations, facilitating the creation of a comprehensive dataset for further analysis.

3CSim: CARLA Corner Case Simulation for Control Assessment in Autonomous Driving

TL;DR

The paper tackles the challenge of evaluating autonomous driving systems under rare and cognitively demanding scenarios. It introduces 3CSim, a CARLA-based framework that enables deterministic, repeatable simulations for both assessing control performance and generating corner-case datasets. A three-part taxonomy—state anomalies, behavior anomalies, and evidence-based anomalies—maps 32 implemented corner-case scenarios with adjustable weather, timing, and traffic density. The work advances safety validation by providing a structured, extensible platform for systematic testing of AD control under non-standard conditions, with future work aimed at expanding scenarios and developing specialized evaluation metrics.

Abstract

We present the CARLA corner case simulation (3CSim) for evaluating autonomous driving (AD) systems within the CARLA simulator. This framework is designed to address the limitations of traditional AD model training by focusing on non-standard, rare, and cognitively challenging scenarios. These corner cases are crucial for ensuring vehicle safety and reliability, as they test advanced control capabilities under unusual conditions. Our approach introduces a taxonomy of corner cases categorized into state anomalies, behavior anomalies, and evidence-based anomalies. We implement 32 unique corner cases with adjustable parameters, including 9 predefined weather conditions, timing, and traffic density. The framework enables repeatable and modifiable scenario evaluations, facilitating the creation of a comprehensive dataset for further analysis.
Paper Structure (10 sections, 8 figures)

This paper contains 10 sections, 8 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Synthesizing Corner Cases
  4. Corner Case Taxonomies
  5. CARLA Corner Case Simulation
  6. Taxonomy of Corner Cases
  7. State Anomaly
  8. Behavior Anomaly
  9. Evidence-Based Anomaly
  10. Conclusion

Figures (8)

  • Figure 1: A wide range of corner cases that can arise in real-world traffic scenarios.
  • Figure 2: Overview of the 3CSim framework for AD system evaluation, including a) input configuration, b) corner case-triggered simulation, and c) output as either scenario assessment or data for further analysis.
  • Figure 3: A taxonomy of corner cases categorized into state anomalies, behavior anomalies, and evidence-based anomalies.
  • Figure 4: A STOP sign integrated into an advertisement.
  • Figure 5: A pedestrian wearing T-shirt displaying the traffic sign STOP.
  • ...and 3 more figures