Generating realistic and diverse pedestrian background flows is critical for numerous downstream applications, ranging from the training and validation of autonomous driving systems to the simulation of mobile communication networks. While recent diffusion-based models achieve state-of-the-art accuracy, they suffer from prohibitive inference latency, lack of physical consistency, and an inability to generalize across heterogeneous datasets, rendering them impractical for industrial Hardware-in-the-Loop testing. To address these challenges, we propose Real-time Adaptive Physics-Informed Diffusion (RAPID), a unified framework explicitly designed to balance high-fidelity generation with strict real-time constraints. First, we introduce a Canonical Representation Module that harmonizes diverse datasets via coordinate-invariant encoding and adaptive modality imputation, enabling unified training across varying scene scales. Second, we propose a Map Context Encoder that decouples computationally expensive map perception from the iterative denoising loop using cached latent embeddings. Third, a Physics-Informed Implicit Sampler integrates Social Force Model gradients as directional priors, encouraging physical consistency (e.g., collision avoidance). Extensive experiments on five heterogeneous benchmarks demonstrate that RAPID establishes a new state-of-the-art balance between fidelity and safety. Notably, it is the only framework capable of operating consistently below the 30 ms industrial threshold, maintaining almost constant inference latency regardless of crowd density. The system exhibits precise controllability over agent behaviors and has been successfully deployed in a production-grade autonomous-driving simulation platform as a core digital twin kernel for large-scale autonomous driving validation. The code is publicly available at https://github.com/tsinghua-fib-lab/RAPID.
The accuracy of consumer-grade inertial navigation is bottlenecked by the stochastic noise of Micro-Electro-Mechanical Systems (MEMS). Traditional deterministic neural architectures often succumb to ``estimation jittering,''sacrificing high-frequency kinematic fidelity for numerical stability. We propose PedestrianDiffusion, a multimodal spectral-domain generative framework reformulating dense 6D state estimation as a continuous conditional denoising process. By operating in the frequency domain, our formulation bounds the spectral covariance, acting as a mathematical preconditioner to stabilize the reverse diffusion trajectory. Furthermore, we introduce a zero-shot semantic conditioning mechanism leveraging vision-language embeddings as categorical priors to generalize across heterogeneous sensor noise profiles. To address the computational intractability of generative tracking, we deploy a single-step deterministic probability flow ODE solver ($T=1$). This yields high-capacity asynchronous batch trajectory refinement, establishing the viability of generative architectures for asynchronous batch trajectory refinement on edge hardware. Extensive evaluations on the OxIOD, RIDI, RoNIN, and TLIO benchmarks demonstrate that PedestrianDiffusion achieves state-of-the-art performance, exhibiting unprecedented robustness to impulse perturbations and coupled 6D kinematic drift. This work provides a rigorous algorithmic blueprint for next-generation Neural Inertial Measurement Units (N-IMUs).
Accurate and computationally efficient traffic prediction remains a fundamental challenge for transportation systems, as microscopic simulators are often too expensive for large-scale applications. This paper addresses this limitation by proposing a physics-consistent surrogate modeling framework, the Graph Sequential Physics-Informed Neural Network (GSPINN). The approach integrates graph-based spatial representation with sequence-aware path aggregation to model trip travel time. It introduces a physics-informed learning formulation that encourages monotonic relationships between travel time and key traffic variables through input-output gradient constraints. To assess robustness across varying traffic conditions, the framework is applied to four heterogeneous road networks, each characterized by distinct topology, demand patterns, and control regimes. The results show consistent predictive performance and stable behavioral properties across all settings. Complementary SHAP-based interpretability further indicates that the model captures network-specific feature dependencies in each case, providing evidence that it adapts to local traffic dynamics rather than overfitting to a single environment. In addition to accuracy and reliability, the proposed surrogate provides substantial computational advantages at inference time, achieving speed-ups of 3x to 55x. This work therefore shows that embedding physically meaningful structure into learning objectives is an effective strategy for traffic surrogate modeling, yielding models that maintain competitive predictive accuracy while substantially improving directional behavioral consistency.
Edge-deployed systems such as autonomous robots, AR/XR devices, and emergency-response handhelds require accurate indoor topological representations, yet existing sensor-based and expert-curated mapping methods are impractical for crowd-sourced, resource-constrained deployment. Additionally, current large language model (LLM) approaches to indoor topology inference lack a systematic framework for evaluating the factors that govern performance. We present Lang2Graph, an experimental framework for indoor topological graph inference from natural-language navigational instructions that isolates four governing factors: instruction structure, metadata clarity, prompting strategy, and model size and reasoning capability. We propose the Independent Prompt Executor (IPE), a prompting strategy that decomposes graph construction into independent per-instruction reasoning steps, preventing error propagation. To support factor-level evaluation, we develop a fully synthetic dataset and an augmented benchmark (R2R-AUG) covering idealized and real-world conditions. Our evaluation across multiple model families show that structured instructions, clear metadata, and IPE improve precision, recall, and F1 by 44%, 45%, and 47%, respectively. Reasoning-aligned open-source models of moderate scale (14B parameters) outperform larger proprietary models on the most challenging instruction categories, indicating that reliable indoor topology inference is achievable without cloud-scale dependencies and establishes a viable path toward on-device edge deployment.
Multisource track association (MTA) is a key measurement-fusion problem in maritime situational awareness, where tracks reported by heterogeneous sensors must be correctly matched under asynchronous sampling, sensor noise, and complex target maneuvers. Existing deep learning-based methods have improved association accuracy by learning discriminative track representations. However, many of them encode two tracks separately and compare only pooled track-level embeddings, so the fine-grained interactions between individual track points are introduced only implicitly. This may limit their ability to distinguish targets with similar motion patterns in dense or ambiguous scenarios. To address this issue, this paper proposes a multilevel interactive Siamese graph neural network, termed (MIS-GNN). First, each track is represented as a graph and encoded by a weight-sharing Siamese GraphSAGE encoder to preserve local spatiotemporal topology. Then, a global interaction layer based on cross-track self-attention is introduced to enable node-level message passing between two track graphs, providing a soft alignment mechanism for asynchronous observations. Finally, graph-to-node cross-attention and multiscale graph-to-graph similarity measurements are jointly employed to evaluate track similarity from complementary granularities. Experiments on the public MTAD benchmark demonstrate that MIS-GNN achieves an average F1-score of 0.9264, outperforming state-of-the-art comparison methods. Additional sensitivity, significance, and efficiency analyses further verify the robustness and deployment feasibility of the proposed method.
Root cause localization is critical for ensuring service reliability in cloud-edge collaborative microservice systems. In practical scenarios, multiple microservice systems are often hybrid-deployed on shared infrastructure, which poses three challenges for existing methods. First, concurrent systems generate substantial metric noise that interferes with anomaly detection. Second, the hierarchical dependencies spanning cloud, edge, and terminal layers cannot be accurately represented in Euclidean space. Third, gateway services that aggregate traffic from multiple systems exhibit amplified anomaly signals, leading to systematic false alarms. To address these issues, we propose confidence gated agent root cause localization (CGARCL), a framework that integrates hyperbolic geometry with confidence gated agent reasoning. CGARCL consists of three components. The direction constrained budgeted anomaly detection method incorporates baseline robust scoring and temporal continuity constraints to extract high-quality candidate anomalous nodes from noisy metrics. The hyperbolic constrained spatio-temporal graph attention network employs Poincar’e ball mapping and center-based topology aggregation to accurately encode hierarchical service dependencies and generate initial root cause rankings. The confidence gated reranking agent is activated when the score gap between the top two candidates is small or the top-ranked node matches a victim-prone pattern. It then performs structured prompt reasoning to suppress false alarms and produce a refined ranking. Experiments on three cloud-edge collaborative microservices datasets demonstrate that CGARCL achieves ACC@1 of 62.1%, 70.6%, and 73.4%, outperforming the second best approach by 19.1%, 11.4%, and 9.8%.