Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Safety-Critical Traffic Simulation with Adversarial Transfer of Driving Intentions

Zherui Huang, Xing Gao, Guanjie Zheng, Licheng Wen, Xuemeng Yang, Xiao Sun

TL;DR

Safety-critical traffic scenario generation is essential for robust autonomous driving but real accident data are scarce. This paper presents IntSim, a two-stage framework that decouples adversarial driving intentions from motion planning: first, it solves a constrained optimization to identify adversarial goals for the OV, then it uses a goal-conditioned trajectory predictor to realize realistic OV and BV motions that steer the AV toward collision. The approach leverages large-scale real-world data and specialized trajectory models to enable efficient exploration of rare events while preserving realism. Extensive experiments on nuScenes and Waymo demonstrate that IntSim achieves state-of-the-art performance in generating adversarial yet realistic safety-critical scenarios and that planners trained with IntSim data exhibit improved safety in challenging near-collision cases.

Abstract

Traffic simulation, complementing real-world data with a long-tail distribution, allows for effective evaluation and enhancement of the ability of autonomous vehicles to handle accident-prone scenarios. Simulating such safety-critical scenarios is nontrivial, however, from log data that are typically regular scenarios, especially in consideration of dynamic adversarial interactions between the future motions of autonomous vehicles and surrounding traffic participants. To address it, this paper proposes an innovative and efficient strategy, termed IntSim, that explicitly decouples the driving intentions of surrounding actors from their motion planning for realistic and efficient safety-critical simulation. We formulate the adversarial transfer of driving intention as an optimization problem, facilitating extensive exploration of diverse attack behaviors and efficient solution convergence. Simultaneously, intention-conditioned motion planning benefits from powerful deep models and large-scale real-world data, permitting the simulation of realistic motion behaviors for actors. Specially, through adapting driving intentions based on environments, IntSim facilitates the flexible realization of dynamic adversarial interactions with autonomous vehicles. Finally, extensive open-loop and closed-loop experiments on real-world datasets, including nuScenes and Waymo, demonstrate that the proposed IntSim achieves state-of-the-art performance in simulating realistic safety-critical scenarios and further improves planners in handling such scenarios.

Safety-Critical Traffic Simulation with Adversarial Transfer of Driving Intentions

TL;DR

Safety-critical traffic scenario generation is essential for robust autonomous driving but real accident data are scarce. This paper presents IntSim, a two-stage framework that decouples adversarial driving intentions from motion planning: first, it solves a constrained optimization to identify adversarial goals for the OV, then it uses a goal-conditioned trajectory predictor to realize realistic OV and BV motions that steer the AV toward collision. The approach leverages large-scale real-world data and specialized trajectory models to enable efficient exploration of rare events while preserving realism. Extensive experiments on nuScenes and Waymo demonstrate that IntSim achieves state-of-the-art performance in generating adversarial yet realistic safety-critical scenarios and that planners trained with IntSim data exhibit improved safety in challenging near-collision cases.

Abstract

Traffic simulation, complementing real-world data with a long-tail distribution, allows for effective evaluation and enhancement of the ability of autonomous vehicles to handle accident-prone scenarios. Simulating such safety-critical scenarios is nontrivial, however, from log data that are typically regular scenarios, especially in consideration of dynamic adversarial interactions between the future motions of autonomous vehicles and surrounding traffic participants. To address it, this paper proposes an innovative and efficient strategy, termed IntSim, that explicitly decouples the driving intentions of surrounding actors from their motion planning for realistic and efficient safety-critical simulation. We formulate the adversarial transfer of driving intention as an optimization problem, facilitating extensive exploration of diverse attack behaviors and efficient solution convergence. Simultaneously, intention-conditioned motion planning benefits from powerful deep models and large-scale real-world data, permitting the simulation of realistic motion behaviors for actors. Specially, through adapting driving intentions based on environments, IntSim facilitates the flexible realization of dynamic adversarial interactions with autonomous vehicles. Finally, extensive open-loop and closed-loop experiments on real-world datasets, including nuScenes and Waymo, demonstrate that the proposed IntSim achieves state-of-the-art performance in simulating realistic safety-critical scenarios and further improves planners in handling such scenarios.

Paper Structure

This paper contains 12 sections, 6 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Problem Definition
  5. Overview
  6. Trajectory Model
  7. Safety-critical Scenario Generation
  8. Experiment
  9. Evaluation for Generated Scenarios
  10. Benifit Planners via Closed-loop Training
  11. Ablation Study
  12. Conclusion and Future Work

Figures (3)

  • Figure 1: Illustration of the framework of IntSim. IntSim employs a decoupling strategy to realize the adversariality and realism of the generated scenarios. First, adversarial intention transfer is formulated as a constrained optimization problem in order to find the best solution to attack the autonomous vehicle directly. Subsequently, guided by the adversarial intention, IntSim leverages a motion prediction model to generate realistic trajectories.
  • Figure 2: Visual comparison of generated scenarios on nuScenes and Waymo datasets. AV, OV, and BV are colored in green, red, and blue, respectively.
  • Figure 3: Qualitative results of ablation study on the motion planning model and the optimization problem. AV, OV, and BV are colored in green, red, and blue, respectively.