In recent years， vision-language-action （VLA） models have attracted significant attention in the field of embodied intelligence. As model scale continues to grow， their ability to generalize across complex tasks has steadily improved. However， such performance improvements rely heavily on the availability of large-scale， high-quality training data. Unlike natural language processing and computer vision， which can directly leverage massive internet data， data collection in embodied intelligence typically involves physical interactions between real robots and their environments， leading to high collection costs and complex acquisition processes. Efficiently obtaining， processing， and organizing such data has therefore become a critical challenge for advancing embodied intelligence. To address this issue， this paper provides a systematic review of data collection and processing methods in embodied intelligence. First， we summarize the major data acquisition paradigms from the perspective of data sources and collection strategies， and analyze their characteristics and limitations in terms of data quality， scalability， and collection cost. Second， we present a standardized processing pipeline for embodied intelligence datasets， focusing on key technical components such as action representation alignment， multimodal temporal synchronization， language semantic normalization， and data quality control. Finally， we discuss the evolving data ecosystem in embodied intelligence， highlighting current challenges and potential future directions. The analysis presented in this paper aims to provide insights for dataset construction and large-scale robot learning research in embodied intelligence.
To meet the demands for comprehensive three-dimensional coverage and massive connectivity in sixth generation of communication system （6G）， establishing a space-air-ground integrated networks （SAGIN） has become a crucial development direction. However， both standalone radio frequency （RF） and free-space optical （FSO） communication technologies have inherent limitations， making it challenging for either to independently fulfill the future network’s comprehensive requirements for ultra-high speed， ultra-high reliability， and wide-area dynamic access. Against this backdrop， integrating the complementary advantages of RF and FSO communications to build an intelligent and cooperative hybrid RF/FSO transmission network for SAGIN has become a key pathway to overcoming current technological bottlenecks. This paper provides a systematic review of domestic and international research advancements in this field， proposing an integrated optical and RF for cognitive software defined network architecture for SAGIN. It focuses on channel modeling methodologies for RF and FSO links applicable to heterogeneous space-air-ground environments， whilst conducting an in-depth analysis of core challenges including high-dynamic link precision alignment， intelligent allocation of heterogeneous resources， and robust transmission under extreme conditions. The paper then elaborates on key enabling technologies such as hybrid RF/FSO beam tracking， adaptive RF/FSO switching， parallel collaborative transmission， and scenario-specific link selection. Future research trends are also outlined， encompassing deep integration of intelligent algorithms， enhancement of cross-domain disturbance-resistant transmission， and holistic system performance optimization. Studies demonstrate that hybrid RF/FSO technology can significantly improve the overall performance of SAGIN. Nevertheless， its path towards large-scale application necessitates further in-depth research on cross-layer coordination mechanisms， dynamic resource management， and system-level performance evaluation.
Bi-temporal optical remote sensing image change detection constitutes a pivotal domain within the broader field of Earth observation,aimed at systematically quantifying terrestrial surface dynamics.By conducting comparative analyses of co-registered imagery acquired over identical geographical coordinates at distinct temporal intervals,this methodology facilitates critical applications ranging from urban expansion monitoring and resource management to disaster damage assessment.The exponential expansion of remote sensing data,coupled with the precipitous maturation of deep learning paradigms,has instigated a transformative era for this discipline.Consequently,the field is witnessing a phase of rapid algorithmic iteration and profound evolutionary growth,significantly enhancing the capability to interpret complex spatiotemporal patterns.Against this backdrop,this manuscript employs a comprehensive chronological framework to systematically represent deep learning-based change detection architectures established over the past two decades.Complementing this survey,it rigorously conducts a comparative analysis,explicitly evaluating both the detection accuracy and computational efficiency of these state-of-the-art methodologies across mainstream benchmark datasets.Beyond mere algorithmic review,the paper consolidates widely utilized public datasets and essential evaluation metrics,thereby providing a standardized reference for benchmarking model performance.Furthermore,this study structurally deconstructs the comprehensive change detection pipeline into its fundamental components.Subsequently,the specific technological advancements and methodological innovations driving the evolution of each critical stage are scrutinized in granular detail to illustrate the workflow's maturation.Ultimately,prospective research frontiers are delineated to forecast the field's developmental trajectory.This outlook aims to serve as a roadmap,offering essential reference and guidance to steer subsequent investigations and foster continued innovation within the domain.
Medical imaging,pathology,and genomics respectively provide information on tumor spatial-morphological phenotypes,histopathological architecture,and molecular mechanisms.Single-modal approaches are constrained by scale discrepancies,sampling biases,and cross-center domain shifts,limiting their support for clinical decision-making.For precision oncology,imaging-pathology-genomics integration aims at semantical alignment and consistency validation among macroscopic imaging,microscopic histological,and mechanistic molecular evidence.This review systematically examines the field via fusion methodologies and clinical applications.We discuss the clinical advantages of multimodal integration,summarize key fusion paradigms,and emphasize its clinical necessity.In applications,focusing on completing the clinical evidence chain,we summarize its advantages in differential diagnosis,molecular subtyping,surgical planning,treatment response stratification,and systematic decision output,highlighting how integration turns predictive results into verifiable,actionable clinical decisions via cross-modal validation.Finally,we discuss emerging trends:spatial omics with multi-region sampling,longitudinal tumor evolution modeling,multimodal foundation models,and multi-center collaborative validation.We propose clinical translation recommendations and a utility evaluation system,offering a roadmap for next-generation intelligent multimodal systems in precision oncology.
With breakthroughs in deep generative technologies such as diffusion models,visual generation models have achieved significant leaps in generation quality and semantic consistency,finding extensive applications in fields like artistic creation and industrial design.However,the powerful generative capability has also triggered severe content safety risks.Malicious users can induce models to generate pornographic,violent,or copyright-infringing images,posing an urgent need for the safety governance of generative AI.This paper provides a systematic review that focuses on two core adversarial tasks of T2I models:(1)Jailbreak attacks,which aim to induce models to breach safety guardrails;(2)Concept erasure,which aims to eliminate internal risk knowledge from the models.First,we establish a taxonomy of jailbreak attacks.By analyzing them across four dimensions:Technical category,perturbation strategy,query type,and adversary knowledge,we reveal the evolutionary trend of attack methods shifting from feature-space perturbations to semantic-space reasoning.Second,regarding risk governance,this paper delves into concept erasure technologies,comparatively analyzing three mainstream technical routes:Model fine-tuning,model editing,and inference guidance.We elucidate the trade-offs among erasure effectiveness,computational efficiency,and the preservation of general generation capabilities.Finally,we summarize the commonly used benchmark datasets in this field and identify the current challenges and future directions regarding adversarial robustness and multi-concept joint governance,aiming to provide theoretical references and technical guidance for building safe and controllable T2I systems.
