High-definition (HD) maps provide precise environmental information essential for prediction and planning in autonomous driving (AD) systems. Due to the high cost of labeling and maintenance, recent research has turned to online HD map construction using onboard sensor data, offering wider coverage and more timely updates for autonomous vehicles (AVs). However, the robustness of online map construction under adversarial conditions remains underexplored. In this paper, we present a systematic vulnerability analysis of online map construction models, which reveals that these models exhibit an inherent bias toward predicting symmetric road structures. In asymmetric scenes like forks or merges, this bias often causes the model to mistakenly predict a straight boundary that mirrors the opposite side. We demonstrate that this vulnerability persists in the real-world and can be reliably triggered by obstruction or targeted interference. Leveraging this vulnerability, we propose a novel two-stage attack framework capable of manipulating online constructed maps. First, our method identifies vulnerable asymmetric scenes along the victim AV's potential route. Then, we optimize the location and pattern of camera-blinding attacks and adversarial patch attacks. Evaluations on a public AD dataset demonstrate that our attacks can degrade mapping accuracy by up to 9.9% in average precision, render up to 44% of targeted routes unreachable, and increase unsafe planned trajectory rates—colliding with real-world road boundaries—by up to 27%. These attacks are also validated on a real-world testbed vehicle. We further analyze root causes of the symmetry bias, attributing them to training data imbalance, model architecture, and map element representation. Based on these findings, we propose asymmetric data fine-tuning as a targeted defense, which significantly improves model robustness. To the best of our knowledge, this study presents the first vulnerability assessment of online map construction models and introduces the first digital and physical attack against them.

Monitoring the Mean Opinion Score (MOS) of video conferencing is critical for Internet Service Providers (ISPs) to ensure user satisfaction. However, a significant technical challenge arises: MOS is a subjective measure, while ISPs primarily rely on deep packet inspection (DPI) data for performance monitoring, making direct mapping between MOS and DPI data nearly impossible. To address this gap, we develop DePI-MOSE, a novel solution that leverages sub-application-level video context clues to build machine-learning models. By inferring the type of end devices and identifying the motion level within video content during a conference session, DePI-MOSE can estimate MOS values from DPI data accurately. We implemented and tested DePI-MOSE in a real-world ISP network, and experimental results show that DePI-MOSE is more accurate than state-of-the-art methods. We also built a network resource management platform for ISPs to dynamically adjust users' network resources by precisely monitoring users' video conferencing QoE.

Multi-modal fusion has shown initial promising results for object detection of autonomous driving perception. However, many existing fusion schemes do not consider the quality of each fusion input and may suffer from adverse conditions on one or more sensors. While predictive uncertainty has been applied to characterize single-modal object detection performance at run time, incorporating uncertainties into the multi-modal fusion still lacks effective solutions due primarily to the uncertainty’s cross-modal incomparability and distinct sensitivities to various adverse conditions. To fill this gap, this paper proposes Uncertainty-Encoded Mixture-of-Experts (UMoE) that explicitly incorporates single-modal uncertainties into LiDAR-camera fusion. UMoE uses individual expert network to process each sensor’s detection result together with encoded uncertainty. Then, the expert networks’ outputs are analyzed by a gating network to determine the fusion weights. The proposed UMoE module can be integrated into any proposal fusion pipeline. Evaluation shows that UMoE achieves a maximum of 10.67%, 3.17%, and 5.40% performance gain compared with the state-of-the-art proposal-level multi-modal object detectors under extreme weather, adversarial, and blinding attack scenarios.

Discretionary lane change (DLC) is a basic but complex maneuver in driving, which aims at reaching a faster speed or better driving conditions, e.g., further line of sight or better ride quality. Although modeling DLC decision-making has been studied for years, the impact of human factors, which is crucial in accurately modelling human DLC decision-making strategies, is largely ignored in the existing literature. In this paper, we integrate the human factors that are represented by driving styles to design a new DLC decision-making model. Specifically, our proposed model takes not only the contextual traffic information but also the driving styles of surrounding vehicles into consideration and makes lane-change/keep decisions. Moreover, the model can imitate human drivers’ decision-making maneuvers by learning the driving style of the ego vehicle. Our evaluation results show that the proposed model captures the human decision-making strategies and imitates human drivers’ lane-change maneuvers, which can achieve 98.66% prediction accuracy. Moreover, we also analyze the lane-change impact of our model compared with human drivers in terms of improving the safety and speed of traffic.

An Autonomous Driving System (ADS) uses a plethora of sensors and many deep learning based tasks to aid its perception, prediction, motion planning, and vehicle control. To ensure road safety, those tasks should be synchronized and use the latest sensing data, which is challenging since 1) different sensors have different sensing periods, 2) the tasks are interdependent, 3) computing resource is limited. This work is the first that uses Age of Information (AoI) as the performance metric for task scheduling in an ADS. We show that minimizing AoI is equivalent to jointly minimizing the response time and maximizing the throughput. We formally formulate the AoI-centric task scheduling problem. To derive practical scheduling solutions, we extend the formulation and formulate the optimal AoI-centric periodic scheduling problem with a given cycle. A reinforcement learning-based solution is designed accordingly. With experiments simulated according to the Apollo driving system, we compare the scheduling performance of the AoI-centric task scheduling with Apollo’s schedulers from the perspective of AoI, throughput, and worst case response time. The experiment results show that the maximum AoI in the proposed scheduling solution with 4 cores is lower than that in Apollo’s schedulers with 8 cores.

The success of machine learning in domains such as computer vision and computer games has triggered a surge of interest in applying machine learning in computer networks. This paper tries to answer a broadly-debated question: can we improve the performance of intradomain routing, one of the most fundamental blocks in the Internet, with reinforcement learning (RL)? Due to the complex network traffic conditions and the large action space in routing, it is difficult to give a definite answer for existing RL-based routing solutions. To gain an in-depth understanding on the challenges of RL-based routing, we systematically classify different RL-based routing solutions and investigate the performance of several representative approaches, in terms of scalability, stability, robustness, and convergence. With the lessons learned in evaluating various RL-based routing solutions, we propose two methods, called supervised Q-network routing (SQR) and discrete link weight-based routing (DLWR), which boost the performance of RL-based routing and outperform the de facto shortest path intradomain routing.

It is envisioned that many different types of applications, each requesting a different set of network functions with different quality of service requirements, will run on 5G networks. To accommodate such requests, network slicing, which dynamically creates virtual networks for different types of applications, is proposed in the literature and IETF. A critical issue for the success of network slicing is to determine the amount of resources for a slice to ensure the required quality of service, i.e., the relationship among traffic demand, resource, and delay. This paper aims to tackle this problem. We first study the delay bound for a given traffic distribution and amount of resources using Stochastic Network Calculus (SNC). Then we propose a solution to derive the amount of resources that should be allocated to meet a given delay bound for a given traffic distribution. Extensive simulations show that given the derived amount of resources, the experienced end-to-end delay is less than the given delay bound. However, if only 90% of the derived amount of resources is given, the experienced end-to-end delay is much higher than the allowed delay bound. This demonstrates that the derived amount of resources is close to the minimum amount of resources for ensuring quality of service. The work provides a useful tool for network slice tenants to decide the amount of resources to request from physical network providers.

Detecting anomalous Border Gateway Protocol (BGP) traffic is significantly important in improving both security and robustness of the Internet. Existing solutions apply classic classifiers to make real-time decision based on the traffic features of present moment. However, due to the frequently happening burst and noise in dynamic Internet traffic, the decision based on short-term features is not reliable. To address this problem, we propose MS-LSTM, a multi-scale Long Short-Term Memory (LSTM) model to consider the Internet flow as a multi-dimensional time sequence and learn the traffic pattern from historical features in a sliding time window. In addition, we find that adopting different time scale to preprocess the traffic flow has great impact on the performance of all classifiers. In this paper, comprehensive experiments are conducted and the results show that a proper time scale can improve about 10% accuracy of LSTM as well as all conventional machine learning methods. Particularly, MS-LSTM with optimal time scale 8 can achieve 99.5% accuracy in the best case.