Deep neural networks (DNNs) are found to be vulnerable against adversarial examples, which carefully crafted inputs with a small magnitude of perturbation aiming induce arbitrarily incorrect predictions. Recent studies show that examples can pose threat real-world security-critical applications: "physical Stop Sign" synthesized such the autonomous driving cars will misrecognize it as others (e.g., speed limit sign). However, these image-space cannot easily alter 3D scans widely equipped...

Validating the safety and performance of an autonomous vehicle (AV) requires benchmarking on real-world driving logs. However, typical logs contain mostly uneventful scenarios with minimal interactions between road users. Identifying interactive in enables curation datasets that amplify critical signals provide a more accurate assessment AV's performance. In this paper, we present novel metric identifies by measuring surprise potential others. First, identify three dimensions design space to...

Autonomous Vehicles (AVs) increasingly use LiDAR-based object detection systems to perceive other vehicles and pedestrians on the road. While existing attacks autonomous driving architectures focus lowering confidence score of AV models induce obstacle misdetection, our research discovers how leverage laser-based spoofing techniques selectively remove LiDAR point cloud data genuine obstacles at sensor level before being used as input perception. The ablation this critical information causes...

It is a basic task in Brillouin distributed fiber sensors to extract the peak frequency of scattering spectrum, since shift gives information on temperature and strain changes. Because high-level noise, quadratic fitting often used data processing. Formulas dependence minimum detectable (BFS) signal-to-noise ratio (SNR) step have been presented publications, but different expressions. A detailed deduction new formulas BFS variance its average given this paper, showing especially their...

Trajectory prediction using deep neural networks (DNNs) is an essential component of autonomous driving (AD) systems. However, these methods are vulnerable to adversarial attacks, leading serious consequences such as collisions. In this work, we identify two key ingredients defend trajectory models against attacks including (1) designing effective training and (2) adding domain-specific data augmentation mitigate the performance degradation on clean data. We demonstrate that our method able...

Polarization fading is a phenomenon observed often in BOTDR distributed sensors, which greatly reduces signal-to-noise ratio of the detected signal. We proposed scheme based on injecting optical pulse probes with orthogonal polarization states, generated by delay Mach-Zehnder interferometer (MZI) composed two beam splitters (PBS). The principle analyzed and effect reducing demonstrated experimentally. method uses simple passive components suitable for practical applications.

To enable safe and reliable decision-making, autonomous vehicles (AVs) feed sensor data to perception algorithms understand the environment. Sensor fusion with multi-frame tracking is becoming increasingly popular for detecting 3D objects. Thus, in this work, we perform an analysis of camera-LiDAR fusion, AV context, under LiDAR spoofing attacks. Recently, LiDAR-only was shown vulnerable attacks; however, demonstrate these attacks are not capable disrupting fusion. We then define a novel,...

Compressive sensing theory have been proposed in the field of radar for target detection. The challenge compressive applied to passive bistatic lies high computational complexity aggravated by transmitted signal dependent time-varying sparse model. So, we propose a transmitted-signal-free and time-invariant model based on orthogonal frequency division multiplexing waveforms. We first generate using surveillance pilot information only, exploiting sparsity scene including only few targets...

Realistic and controllable traffic simulation is a core capability that necessary to accelerate autonomous vehicle (AV) development. However, current approaches for controlling learning-based models require significant domain expertise are difficult practitioners use. To remedy this, we present CTG++, scene-level conditional diffusion model can be guided by language instructions. Developing this requires tackling two challenges: the need realistic backbone, an effective method interface with...

A phase-sensitive optical time domain reflectometer (φ-OTDR) based on a 120°-phase-difference Michelson interferometer is proposed. The with arm difference of 4m used to test the phase between Rayleigh scattering from two sections fiber. new demodulation method called inverse transmission matrix scheme utilized demodulate distributed backward along long experimental results show that φ-OTDR can detect 3km fiber, and acoustic signal within whole human hearing range 20 Hz–20 kHz reproduced...

Performances of centroid analysis (CA) used for extracting Brillouin frequency shift (BFS) from noisy signals are studied in this paper. The variance extracted BFS, i.e., the minimum detectable is deduced as a function signal-to-noise ratio, step, linewidth, and data window analysis. It found theoretically that both averaged BFS its susceptible to deviation center real-Brillouin central frequency, termed (DWD). analyzed results verified by experiments simulation, showing good agreement with...

An important paradigm in 3D object detection is the use of multiple modalities to enhance accuracy both normal and challenging conditions, particularly for long-tail scenarios. To address this, recent studies have explored two directions adaptive approaches: MoE-based fusion, which struggles with uncertainties arising from distinct configurations, late fusion output-level relies on separate pipelines limits comprehensive understanding. In this work, we introduce Cocoon, an object-...

Abstract Although some existing sparse representation (SR) methods are robust for target detection in passive bistatic radar (PBR), they still face the challenges of high computational complexity and poor performance extremely low-signal-to-clutter ratio (SCR) target. So, an average effective subcarrier (AES)-domain approach is investigated this paper. Firstly, AES-based SR model proposed to solve problem complexity, which established by utilizing sparseness orthogonal frequency-division...

With the fast development of large language models (LLMs), LLM-driven Web Agents (Web for short) have obtained tons attention due to their superior capability where LLMs serve as core part making decisions like human brain equipped with multiple web tools actively interact external deployed websites. As uncountable been released and such LLM systems are experiencing rapid drawing closer widespread deployment in our daily lives, an essential pressing question arises: "Are these secure?". In...

Deep neural networks (DNNs) are increasingly integrated into LiDAR (Light Detection and Ranging)-based perception systems for autonomous vehicles (AVs), requiring robust performance under adversarial conditions. We aim to address the challenge of spoofing attacks, where attackers inject fake objects data fool AVs misinterpret their environment make erroneous decisions. However, current defense algorithms predominantly depend on outputs (i.e., bounding boxes) thus face limitations in...

This paper presents a method for semi-supervised classification of polarimetric synthetic aperture radar (PolSAR) sea ice data. The consists two steps. In the first stage, markov random field on region adjacency graph is constructed initial watershed oversegmented result. While in second Wishart distribution model and maximum posterior (MAP) are applied as criterion obtaining optimal classification. Good experimental results less time- consuming obtained when this to PolSAR data sets...

Simulation plays a crucial role in the development of autonomous vehicles (AVs) due to potential risks associated with real-world testing. Although significant progress has been made visual aspects simulators, generating complex behavior among agents remains formidable challenge. It is not only imperative ensure realism scenarios generated but also essential incorporate preferences and conditions facilitate controllable generation for AV training evaluation. Traditional methods, mainly...

With the development of electronics and information technology, requirement people toward resolution images become higher. However, acquired can't meet our needs because different imaging system from a variety factors. And it will cost too much to improve image via using physical hardware. So requires new method solve this problem. Super-resolution is emerged in context. This technique uses software approach enhance based on existing images. Thus has aroused widespread concern when appeared....

We demonstrate a narrow-linewidth laser source for high spatial resolution distributed optical sensing by utilizing the high-order modulation sidebands injection locking. A pair of phase-locked lasers with arbitrary frequency offset from 5 GHz to 50 is generated. Meanwhile, linearized sweep covering range 15 in 6 ms errors 240 kHz linearity also achieved using same scheme, instantaneous linewidth frequency-swept measured be ~2.5 kHz.