Acceleration method for generating perception failure scenarios based on editing Markov process

TL;DR

The paper tackles the challenge of efficiently validating autonomous vehicle perception in underground parking garages by proposing an accelerated generation method that learns interactions between background vehicles (BVs) and the autonomous vehicle (AV) and by editing the Markov process to emphasize safety-critical, perception-relevant states. It formulates the problem as maximizing a perception-failure probability $P(A)$ via optimizing the BV maneuver distribution $P_q$ within a Markov framework, and derives a tractable deep-learning loss $L$ that focuses on critical states to enhance learning efficiency. The authors validate their approach in a Carla–Vissim coupled simulation with Bevfusion as the perception algorithm, showing that training on critical states yields faster convergence and that the resulting intelligent testing environments can yield high-density perception-failure scenarios and improved safety performance after retraining. The work demonstrates a data-driven, scalable path to robustly stress-test perception modules in structurally constrained, low-light settings, enabling safer deployment of AVs in underground parking infrastructures.

Abstract

With the rapid advancement of autonomous driving technology, self-driving cars have become a central focus in the development of future transportation systems. Scenario generation technology has emerged as a crucial tool for testing and verifying the safety performance of autonomous driving systems. Current research in scenario generation primarily focuses on open roads such as highways, with relatively limited studies on underground parking garages. The unique structural constraints, insufficient lighting, and high-density obstacles in underground parking garages impose greater demands on the perception systems, which are critical to autonomous driving technology. This study proposes an accelerated generation method for perception failure scenarios tailored to the underground parking garage environment, aimed at testing and improving the safety performance of autonomous vehicle (AV) perception algorithms in such settings. The method presented in this paper generates an intelligent testing environment with a high density of perception failure scenarios by learning the interactions between background vehicles (BVs) and autonomous vehicles (AVs) within perception failure scenarios. Furthermore, this method edits the Markov process within the perception failure scenario data to increase the density of critical information in the training data, thereby optimizing the learning and generation of perception failure scenarios. A simulation environment for an underground parking garage was developed using the Carla and Vissim platforms, with Bevfusion employed as the perception algorithm for testing. The study demonstrates that this method can generate an intelligent testing environment with a high density of perception failure scenarios and enhance the safety performance of perception algorithms within this experimental setup.

Figures (9)

• Figure 1: Carla simulation environment of the underground parking garage. The model is a 1:1 simulation of the underground parking garage at the Southern University of Science and Technology of Engineering, including lane markings, parking spaces, and traffic signs with semantic information.
• Figure 2: Vissim underground parking garage network. Red boxes are examples of path decision points in Vissim, where vehicles are assigned one of the three yellow paths shown upon passing these points. Black boxes indicate the vehicle entry and exit points; dark blue boxes represent parking spaces; gray areas denote the roads.
• Figure 3: The coupled simulation process with Carla and Vissim.
• Figure 4: AV driving route (highlighted in yellow).
• Figure 5: Loss curve during Bevfusion training.