Autonomous driving systems have witnessed significant development during the past years thanks to advance in machine learning-enabled sensing and decision-making algorithms. One critical challenge for their massive deployment real world is safety evaluation. Most existing are still trained evaluated on naturalistic scenarios collected from daily life or heuristically-generated adversarial ones. However, large population of cars, general, leads an extremely low collision rate, indicating that...

Existing neural network-based autonomous systems are shown to be vulnerable against adversarial attacks, therefore sophisticated evaluation of their robustness is great importance. However, evaluating the under worst-case scenarios based on known attacks not comprehensive, mention that some them even rarely occur in real world. Also, distribution safety-critical data usually multimodal, while most traditional and methods focus a single modality. To solve above challenges, we propose...

Long-tail and rare event problems become crucial when autonomous driving algorithms are applied in the real world. For purpose of evaluating systems challenging settings, we propose a generative framework to create safety-critical scenarios for specific task algorithms. We first represent traffic with series autoregressive building blocks generate diverse by sampling from joint distribution these blocks. then train model as an agent (or generator) search risky scenario parameters given...

Naturalistic driving trajectory generation is crucial for the development of autonomous algorithms. However, most data collected in collision-free scenarios leading to sparsity safety-critical cases. When considering safety, testing algorithms near-miss that rarely show up off-the-shelf datasets and are costly accumulate a vital part evaluation. As remedy, we propose synthesizing framework based on variational Bayesian methods term it as Conditional Multiple Trajectory Synthesizer (CMTS). We...

Generating adversarial scenes that potentially fail autonomous driving systems provides an effective way to improve their robustness. Extending purely data-driven generative models, recent specialized models satisfy additional controllable requirements such as embedding a traffic sign in scene by manipulating patterns <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">implicitly</i> at the neuron level. In this paper, we introduce method...

Active perception describes a broad class of techniques that couple planning and systems to move the robot in way give more information about environment. In most robotic systems, is typically independent motion planning. For example, traditional object detection passive: it operates only on images receives. However, we have chance improve results if allow consume signals collect views maximize quality results. this paper, use reinforcement learning (RL) methods control order obtain quality....

In this paper, we present a method to estimate vehicle's pose and shape from off-board multi-view images. These images are taken monocular cameras with small overlaps. We utilize state-of-the-art Convolutional Neural Networks (CNNs) extract vehicles' semantic keypoints introduce Cross Projection Optimization (CPO) the 3D pose. During iterative CPO process, an adaptive adjustment named Hierarchical Wireframe Constraint (HWC) is implemented shape. Our approach evaluated under both simulated...

In this paper, we propose a navigation algorithm oriented to multi-agent environment. This is expressed as hierarchical framework that contains Hidden Markov Model (HMM) and Deep Reinforcement Learning (DRL) structure. For simplification, term our method Hierarchical Navigation Network (HNRN). high-level architecture, train an HMM evaluate the agents perception obtain score. According score, adaptive control action will be chosen. While in low-level two sub-systems are introduced, one...

As shown by recent studies, machine intelligence-enabled systems are vulnerable to test cases resulting from either adversarial manipulation or natural distribution shifts. This has raised great concerns about deploying learning algorithms for real-world applications, especially in safety-critical domains such as autonomous driving (AD). On the other hand, traditional AD testing on naturalistic scenarios requires hundreds of millions miles due high dimensionality and rareness real world. a...

Safety is a critical concern when deploying reinforcement learning agents for realistic tasks. Recently, safe algorithms have been developed to optimize the agent's performance while avoiding violations of safety constraints. However, few studies addressed nonstationary disturbances in environments, which may cause catastrophic outcomes. In this paper, we propose context-aware (CASRL) method, metal-earning framework realize adaptation non-stationary environments. We use probabilistic latent...

An accurate estimation of the state health (SOH) Li-ion batteries is critical for efficient and safe operation battery-powered systems. Traditional methods SOH estimation, such as Coulomb counting, often struggle with sensitivity to measurement noise time-consuming tests. This study addresses this issue by combining incremental capacity (IC) analysis a novel neural network, Kolmogorov–Arnold Networks (KANs). Fifteen features were extracted from IC curves 2RC equivalent circuit model was used...

In this paper, we propose an enhanced triplet method that improves the encoding process of embeddings by jointly utilizing generative adversarial mechanism and multitasking optimization. We extend our encoder with Generative Adversarial Networks (GANs) softmax loss function. GAN is introduced for increasing generality diversity samples, while reinforcing features about speakers. For simplification, term Multitasking Triplet (MTGAN). Experiment on short utterances demonstrates MTGAN reduces...

Generating multi-vehicle trajectories from existing limited data can provide rich resources for autonomous vehicle development and testing. This paper introduces a trajectory generator (MTG) that encode interaction scenarios (called driving encounters) into an interpretable representation which new encounter are generated by sampling. The MTG consists of bi-directional encoder multi-branch decoder. A disentanglement metric is then developed model analyses comparisons in terms robustness the...

A trustworthy reinforcement learning algorithm should be competent in solving challenging real-world problems, including {robustly} handling uncertainties, satisfying {safety} constraints to avoid catastrophic failures, and {generalizing} unseen scenarios during deployments. This study aims overview these main perspectives of considering its intrinsic vulnerabilities on robustness, safety, generalizability. In particular, we give rigorous formulations, categorize corresponding methodologies,...

With the increment of depth metal mines, energy conservation refrigeration ventilation for deep mines has attracted more attention. This study uses heat current method and artificial neural network to model system establishes systematic flow constraints using driving-drag balance relationship. A global operational optimization is built a Lagrange multiplier minimize pumping power consumption. An iterative algorithm combined with stope proposed by coupling water outlet temperature evaporator...

Different environments pose a great challenge to the outdoor robust visual perception for long-term autonomous driving, and generalization of learning-based algorithms on different is still an open problem. Although monocular depth prediction has been well studied recently, few works focus robustness across environments, e.g. changing illumination seasons, owing lack such multi-environment real-world dataset benchmark. To this end, cross-season benchmark, SeasonDepth, introduced benchmark...

Continuously learning to solve unseen tasks with limited experience has been extensively pursued in meta-learning and continual learning, but restricted assumptions such as accessible task distributions, independently identically distributed tasks, clear delineations. However, real-world physical frequently violate these assumptions, resulting performance degradation. This paper proposes a online model-based reinforcement approach that does not require pre-training task-agnostic problems...

The goal of acoustic (or sound) events detection (AED or SED) is to predict the temporal position target in given audio segments. This task plays a significant role safety monitoring, early warning and other scenarios. However, deficiency data diversity event sources make AED tough issue, especially for prevalent data-driven methods. In this article, we start from analyzing according their time-frequency domain properties, showing that different have scale characteristics. Inspired by...

Evaluating the reliability of intelligent physical systems against rare safety-critical events poses a huge testing burden for real-world applications. Simulation provides useful platform to evaluate extremal risks these before their deployments. Importance Sampling (IS), while proven be powerful rare-event simulation, faces challenges in handling learning-based due black-box nature that fundamentally undermines its efficiency guarantee, which can lead under-estimation without diagnostically...

Crowd monitoring involves tracking and analyzing the behavior of large groups people in large-scale public spaces, such as sports games. In stadiums, understanding audience reactions to games their distribution around facilities is important for ensuring safety security, enhancing game experience, improving crowd management. Recent crowd-crushing incidents (e.g., Kanjuruhan Stadium disaster, Seoul Halloween Stampede) have caused 100+ deaths a single event, calling advancements methods....

Generating multi-vehicle trajectories from existing limited data can provide rich resources for autonomous vehicle development and testing. This paper introduces a trajectory generator (MTG) that encode interaction scenarios (called driving encounters) into an interpretable representation which new encounter are generated by sampling. The MTG consists of bi-directional encoder multi-branch decoder. A disentanglement metric is then developed model analyses comparisons in terms robustness the...