Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Anomaly Detection in Cooperative Vehicle Perception Systems under Imperfect Communication

Ashish Bastola, Hao Wang, Abolfazl Razi

TL;DR

The paper tackles anomaly detection in autonomous driving when vehicles rely on imperfect cooperative perception. It introduces CPAD, a graph-transformer framework that models spatio-temporal correlations and remains effective under communication interruptions. A 15,000-scenario, 90,000-trajectory benchmark is proposed, along with a robust CPAD architecture featuring graph-based spatial fusion and a Transformer-based temporal classifier, plus extensive robustness analyses under node blackouts. The results show CPAD outperforms standard baselines on F1 and AUC and maintains high detection performance despite substantial communication losses, with the dataset and code made publicly available to advance research and practical deployment.

Abstract

Anomaly detection is a critical requirement for ensuring safety in autonomous driving. In this work, we leverage Cooperative Perception to share information across nearby vehicles, enabling more accurate identification and consensus of anomalous behaviors in complex traffic scenarios. To account for the real-world challenge of imperfect communication, we propose a cooperative-perception-based anomaly detection framework (CPAD), which is a robust architecture that remains effective under communication interruptions, thereby facilitating reliable performance even in low-bandwidth settings. Since no multi-agent anomaly detection dataset exists for vehicle trajectories, we introduce 15,000 different scenarios with a 90,000 trajectories benchmark dataset generated through rule-based vehicle dynamics analysis. Empirical results demonstrate that our approach outperforms standard anomaly classification methods in F1-score, AUC and showcase strong robustness to agent connection interruptions.

Anomaly Detection in Cooperative Vehicle Perception Systems under Imperfect Communication

TL;DR

The paper tackles anomaly detection in autonomous driving when vehicles rely on imperfect cooperative perception. It introduces CPAD, a graph-transformer framework that models spatio-temporal correlations and remains effective under communication interruptions. A 15,000-scenario, 90,000-trajectory benchmark is proposed, along with a robust CPAD architecture featuring graph-based spatial fusion and a Transformer-based temporal classifier, plus extensive robustness analyses under node blackouts. The results show CPAD outperforms standard baselines on F1 and AUC and maintains high detection performance despite substantial communication losses, with the dataset and code made publicly available to advance research and practical deployment.

Abstract

Anomaly detection is a critical requirement for ensuring safety in autonomous driving. In this work, we leverage Cooperative Perception to share information across nearby vehicles, enabling more accurate identification and consensus of anomalous behaviors in complex traffic scenarios. To account for the real-world challenge of imperfect communication, we propose a cooperative-perception-based anomaly detection framework (CPAD), which is a robust architecture that remains effective under communication interruptions, thereby facilitating reliable performance even in low-bandwidth settings. Since no multi-agent anomaly detection dataset exists for vehicle trajectories, we introduce 15,000 different scenarios with a 90,000 trajectories benchmark dataset generated through rule-based vehicle dynamics analysis. Empirical results demonstrate that our approach outperforms standard anomaly classification methods in F1-score, AUC and showcase strong robustness to agent connection interruptions.

Paper Structure

This paper contains 20 sections, 13 equations, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Collective information sharing
  4. Graph-based information modeling
  5. Anomaly detection in Communication constrained Scenarios
  6. Problem Formulation
  7. Multi-agent configuration
  8. Rule-based anomaly labeling
  9. Zigzag driving patterns
  10. Sudden Braking
  11. Sudden turns
  12. Frequent lane switching/driving between two lanes
  13. Tailgating/ Driving very close
  14. Algorithm Design
  15. Graph based Spatial sensor fusion
  16. ...and 5 more sections

Figures (6)

  • Figure 1: Demonstrates individual perception of vehicles in 6 agent settings. We observe similar perception but in the form of lidars, lane detectors and side detectors instead of visual graphics for computational efficiency.
  • Figure 2: Anomalies classified using rule-based approach
  • Figure 3: Graph-Transformer Architecture of the Cooperative Perception-Based Anomaly Detection (CPAD) Model for Trajectory Anomaly Detection.
  • Figure 4: Node blackout configurations. Sequential blackout on the left where the node goes out of communication for a certain period of time. Random blackout on the right where the node goes through intermittent connection loss.
  • Figure 5: Confusion Matrix of True vs. Predicted Anomaly Labels
  • ...and 1 more figures