Reinforcement learning is a core technology for modern artificial intelligence, and it has become a workhorse for AI applications ranging from Atrai Game to Connected and Automated Vehicle System (CAV). Therefore, a reliable RL system is the foundation for the security critical applications in AI, which has attracted a concern that is more critical than ever. However, recent studies discover that the interesting attack mode adversarial attack also be effective when targeting neural network policies in the context of reinforcement learning, which has inspired innovative researches in this direction. Hence, in this paper, we give the very first attempt to conduct a comprehensive survey on adversarial attacks in reinforcement learning under AI security. Moreover, we give briefly introduction on the most representative defense technologies against existing adversarial attacks.

Inefficient signal control will not only exaggerate traffic congestion, but also increase the fuel consumption and exhaust emissions. Thus, signal planning is highly important in green transportation. As the Connected vehicle (CV) technology has transformed today’s transportation systems by connecting vehicles and the transportation infrastructure through wireless communication, the CV-based signal control system has seen significant studies recently. Unfortunately, existing signal planning algorithms in use are developed for the signal-intersection, showing low traffic efficiency in the multi-intersection collaborative planning due to ignoring the traffic correlation among the neighboring intersections. In this work, we target the USDOT (U.S. Department of Transportation) sponsored CV-based traffic control system, and implement a multi-intersection traffic network. We model the multi-intersection collaborative signal planning problem as a multi-agent reinforcement learning problem, and present an actor-attention-critic algorithm to improve transportation efficiency and energy efficiency in green transportation, as well as resist congestion attack. Experiment results on the multi-intersection traffic network indicates that 1) compared to the baseline, our approach reduces the total delay by as high as 44.24%; 2) our method transports more vehicles passing the intersections meanwhile reduces the total CO ₂ emissions by 2.40%; 3) under the congestion attack, our approach shows robustness and reduces the total delay by as high as 64.33%.

While autonomous driving systems and intelligent transportation infrastructures become increasingly software-defined and network-connected, ensuring their cybersecurity has become a critical component of traffic safety. Large language models (LLMs) have recently shown promise in automating aspects of penetration testing, yet most existing approaches remain limited to simple, single-step exploits. They struggle to handle complex, multi-stage vulnerabilities that demand precise coordination, contextual reasoning, and knowledge reuse. This is particularly problematic in safety-critical domains, such as autonomous vehicles, where subtle software flaws can cascade across interdependent subsystems. In this work, we present CurriculumPT, a novel LLM-based penetration testing framework specifically designed for the security of intelligent systems. CurriculumPT combines curriculum learning and a multi-agent system to enable LLM agents to progressively acquire and apply exploitation skills across common vulnerabilities and exposures-based tasks. Through a structured progression from simple to complex vulnerabilities, agents build and refine an experience knowledge base that supports generalization to new attack surfaces without requiring model fine-tuning. We evaluate CurriculumPT on 15 real-world vulnerabilities scenarios and demonstrate that it outperforms three state-of-the-art baselines by up to 18 percentage points in exploit success rate, while achieving superior efficiency in execution time and resource usage. Our results confirm that CurriculumPT is capable of autonomous, scalable penetration testing and knowledge transfer, laying the groundwork for intelligent security auditing of modern autonomous driving systems and other cyberphysical transportation platforms.

With the rapid development of wireless communication technology and intelligent mobile devices, unmanned aerial vehicle (UAV) cluster is becoming increasingly popular in both civilian and military applications. Recently, a swarm intelligence-based UAV cluster study, aiming to enable efficient and autonomous collaboration, has drawn lots of interest. However, new security problems may be introduced with such swarm intelligence. In this work, we perform the first detailed security analysis to a kind of flocking-based UAV cluster with 5 policies, an upgrade version of the well-known Boids model. Targeting a realistic threat in a source-to-destination flying task, we design a data spoofing strategy and further perform complete vulnerability analysis. We reveal that such design and implementation are highly vulnerable. After breaking through the authentication of ad hoc on-demand distance vector (AODV) routing protocol by rushing attack, an attacker can masquerade as the first-arrival UAV within a specific scope of destination and generate data spoofing of arrival status to the following UAVs, so as to interfere with their normal flying paths of destination arrival and cause unexpected arrival delays amid urgent tasks. Experiments with detailed analysis from the 5-UAV cluster to the 10-UAV cluster are conducted to show specific feature composition-based attack effect and corresponding average delay. We also discuss promising defense suggestions leveraging the insights from our analysis.

Recent advances in deep learning (DL) have brought tremendous gains in signal modulation classification. However, DL-based classifiers lack transparency and interpretability, which raises concern about model's reliability and hinders the wide deployment in real-word applications. While explainable methods have recently emerged, little has been done to explain the DL-based signal modulation classifiers. In this work, we propose a novel model-agnostic explainer, Model-Agnostic Signal modulation classification Explainer (MASE), which provides explanations for the predictions of black-box modulation classifiers. With the subsequence-based signal interpretable representation and in-distribution local signal sampling, MASE learns a local linear surrogate model to derive a class activation vector, which assigns importance values to the timesteps of signal instance. Besides, the constellation-based explanation visualization is adopted to spotlight the important signal features relevant to model prediction. We furthermore propose the first generic quantitative explanation evaluation framework for signal modulation classification to automatically measure the faithfulness, sensitivity, robustness, and efficiency of explanations. Extensive experiments are conducted on two real-world datasets with four black-box signal modulation classifiers. The quantitative results indicate MASE outperforms two state-of-the-art methods with 44.7% improvement in faithfulness, 30.6% improvement in robustness, and 44.1% decrease in sensitivity. Through qualitative visualizations, we further demonstrate the explanations of MASE are more human interpretable and provide better understanding into the reliability of black-box model decisions.

Pedestrian detection plays a crucial role in autonomous driving by identifying the position, size, orientation, and dynamic features of pedestrians in images or videos, assisting autonomous vehicles in making better decisions and controls. It’s worth noting that the performance of pedestrian detection models largely depends on the quality and diversity of available training data. Current datasets for autonomous driving have limitations in terms of diversity, scale, and quality. In recent years, numerous studies have proposed the use of data augmentation strategies to expand the coverage of datasets, aiming to maximize the utilization of existing training data. However, these data augmentation methods often overlook the diversity of data scenarios. To overcome this challenge, in this paper, we propose a more comprehensive method for data augmentation, based on image descriptions and diffusion models. This method aims to cover a wider range of scene variations, including different weather conditions and lighting situations. We have designed a classifier to select data samples for augmentation, followed by extracting visual features based on image captions and converting them into high-level semantic information as textual descriptions for the corresponding samples. Finally, we utilize diffusion models to generate new variants. Additionally, we have designed three modification patterns to increase diversity in aspects such as weather conditions, lighting, and pedestrian poses within the data. We conducted extensive experiments on the KITTI dataset and in real-world environments, demonstrating that our proposed method significantly enhances the performance of pedestrian detection models in complex scenarios. This meticulous consideration of data augmentation will notably enhance the applicability and robustness of pedestrian detection models in actual autonomous driving scenarios.