City systems are shaped by interactions among infrastructure, human activity, and environmental conditions. Existing earth observation and urban benchmarks are limited in unified modeling and in capturing their joint dynamics. We introduce AgenticCity-Bench, a longitudinal multimodal benchmark for multi-layer unified state modeling and agentic city task reasoning. The benchmark covers 16 cities in 12 countries across six continents from 2000 to 2025, and includes 18 modalities spanning urban function, activity, and environmental state. It organizes heterogeneous data sources around a common spatial reference and calendar index while preserving spatial and temporal coverage, provenance and tile-level availability. On this foundation, AgenticCity-Bench supports urban cross-modal translation together with diverse agentic city tasks involving forecasting, planning and resource analysis. This study provides a unified benchmark for studying the transition from multimodal urban and earth observation to joint understanding and agentic city decision-making. The dataset is available at this https URL.
https://arxiv.org/abs/2610.35293
Quadrotor controller comparisons commonly report a single tracking error, although the resulting ranking also depends on tuning effort, reference information, actuator prediction, optimizer fallback, and command timing. We quantify these effects for cascade PID, hover LQR, boundary-layer sliding-mode control (BL-SMC), linear MPC with LQR fallback (LMPC-F), a preview-enabled LMPC-F, and a geometric $SO(3)$ controller on a common nonlinear plant with rotor gyroscopic coupling, motor lag, and saturation. A primary study (556 episodes at 50~Hz) combines matched twelve-candidate tuning, 400 paired evaluation episodes, and a 48-cell MPC factorial; a confirmation study (1462 episodes) repeats the evaluation on 50 fresh realizations with tuning budgets of 12, 24, and 48 candidates. A matched twelve-candidate tuning protocol changed the mean-RMSE ordering obtained with the fixed baseline settings, moving BL-SMC from last to first; the new ordering recurred on the fresh realizations, and a 48-candidate budget left it unchanged. Reference preview substantially reduced MPC tracking error and, on the figure-eight task, removed fallback activity. On an abrupt step, the outcome depended on the interplay between fallback policy, predictor, and actuator authority. Command timing also changed the outcome: replaying measured command-call delays terminated three of four nominal MPC configurations on a desktop computer and all four on a Raspberry Pi~5, whereas capping the solver at 100 iterations removed every deadline miss in the tested nominal timing profiles. Controller rankings therefore depend not only on tracking accuracy but also on the tuning protocol, reference information, fallback behavior, and implementation timing. We propose a reporting checklist covering tracking accuracy, tuning effort, completion and failure rates, fallback activity, applied-input variation, and command timing.
https://arxiv.org/abs/2610.35165
AI assistants are increasingly asked to reason over a user's own web traces: group chats, community threads, activity logs and revision histories. On CL-bench Life, a benchmark of such real-life context learning, frontier LLMs solve fewer than one task in five. We find that the request is not the specification: over 80% of the benchmark's rubrics share less than 60% of their content words with the final request, as for a rule set two turns earlier or every record that meets the user's criterion, and LLMs answering directly pass only 67% of these implicit rubrics against 81% of the others. We propose QERA, a Question-driven Evidence Reasoning Architecture that asks before it answers. QERA first turns the request into a requirement checklist and typed information needs, filtered by code for faithfulness to the user's words; it then answers from the whole conversation, asks the same questions again with quotes verified by code, and applies a guarded local repair. It needs no training or rubrics and makes about four LLM calls per task. On all 405 tasks with four frontier LLMs, QERA attains the highest solving rate on every LLM (21.0% on average, against 18.2% for the strongest of three inference-time baselines and 14.6% for direct answering) and leads every baseline significantly when pooled across LLMs, while using fewer tokens than PSCI and 4-5% of the tokens of Ctx2Skill. Ablations on GPT-6-Luna show that the gain is settled before the answer, by a fixed answer protocol (+3.8 rubric points) and the request-specific questions (+1.6), and that it is concentrated on implicit rubrics, with checking afterwards refining mainly numeric requirements.
https://arxiv.org/abs/2610.35083
An agent that truly knows and acts on behalf of its user has been a long-standing promise of AI assistants. Computer-use agents bring this promise within reach by capably executing fully specified tasks across apps. However, real-world user requests are often underspecified. To act correctly for the particular user, an agent must infer some missing information from the user's activity history and ask for other information that only the user can provide, without making unsupported assumptions or burdening the user with unnecessary questions. We present iPrefBench, a benchmark for evaluating the infer and ask capabilities of personal mobile-use agents in an executable iOS environment with an interactive simulated user. Grounded in a coherent persona with 3,051 past activity events, iPrefBench comprises 130 underspecified user tasks across 26 apps. Each omitted user-dependent requirement is, by construction, either history-inferable or ask-only. iPrefBench evaluates personalized mobile agents on both whether agents complete tasks correctly for the user (i.e., personalized task completion) and whether they seek user input when necessary (i.e., user-agent interaction). Evaluation of 7 frontier agents reveals a substantial gap between execution and personalization. Agents achieve high action completion scores (up to 91.6%) but low full task success (up to 47.7%) because personalization lags behind. They especially struggle with ask-only information, failing to request enough necessary information. Our findings highlight a challenge for personalized agents in knowing what to infer and when to ask. We hope that iPrefBench can serve as a valuable testbed for building agents that know their user and work with them.
https://arxiv.org/abs/2610.34735
This paper presents S2SND-v2, a unified online and offline speaker diarization framework built on S2SND that improves acoustic feature extraction and the use of the resulting representations. For acoustic feature extraction, we introduce a speaker-adapted w2v-BERT 2.0 front end and compare full and pruned construction routes. To better use these features, overlap-aware activity-query masking improves speaker embedding extraction under overlap, while late-speaker training aligns unknown-speaker training with online inference conditions. Combining these improvements, the pruned system achieves competitive average diarization error rates of 14.52% online and 12.03% offline across eight test sets.
https://arxiv.org/abs/2610.34707
Animals have a remarkable capacity for autonomy---driven by intrinsic motivation---without external sensory cues or extrinsic objectives, yet the computational mechanisms behind this ability largely remain a mystery. We combine whole-brain recordings in the fruit fly (Drosophila melanogaster), which reveal neural representations of spontaneous exploratory decisions more than 10 seconds in advance, with a neuromechanical digital twin to test whether optimizing intrinsic goals yields realistically embodied artificial agents that quantitatively match fly behavior and internal representations of future volitional action during exploration. Our hierarchical policy selects from a discrete library of low-level motor primitives, consistent with ethological theories of behavioral syllables. We train agents without external reward using 20 intrinsic motivation algorithms spanning curiosity, entropy, and competency methods. The agents are evaluated in a virtual tethered-ball assay modeled on the fly experiment, comparing spontaneous locomotor statistics and internal representations and decoding future turn direction from kinematics and neural activity on the same timescales observed in flies. DIAYN, a hierarchical competency algorithm learning discrete policies that maximize behavioral entropy and the separability of induced sensory observations, achieves the strongest joint behavioral and neural prospective coding performance relative to flies. Its learned policies closely capture individual variability in fly exploration strategies, suggesting that this diversity serves to maximize state-space coverage at the population level. DIAYN's latent activity also closely captures the sum-difference rule for interhemispheric encoding of kinematics in biological flies. These results establish hierarchical, intrinsically motivated agents as a powerful framework for reverse-engineering animal autonomy.
https://arxiv.org/abs/2610.34638
Recently, point-supervised temporal action localization has attracted increasing attention due to its low annotation cost (only one frame per action instance) and competitive detection performance. However, since point annotations are less complete than proposal annotations, existing methods mainly rely on multiple instance learning for snippet-level classification, which leads to a performance gap compared to fully supervised models. To bridge this gap, we propose P2P-Former, a novel point-to-proposal Transformer-based framework that effectively connects point-level supervision with proposal-level learning for temporal action localization. In this work, we identify two key challenges in transitioning from point-level supervision to full supervision: (1) how to generate reliable proposals as pseudo ground-truth, and (2) how to effectively integrate point-level cues into the fully supervised paradigm. To address the first challenge, we propose a Point-aware Proposal Generation (PPG) strategy that refines the confidence of proposals produced by a base locator using a point-aware importance estimation function, and then performs optimal matching between point annotations and candidate proposals to select reliable proposals as pseudo ground-truth. To tackle the second challenge, we introduce a Point-guided Semantic Enhancement (PSE) module that aggregates prototypes from point-annotated features and injects them into snippet features through Transformer-based interactions. Moreover, we design a Point-to-Proposal Constraint Loss (PPCL) that enforces each predicted proposal to include the annotated point, thereby encouraging more accurate localization. Extensive experiments on the THUMOS14 and ActivityNet-1.3 datasets demonstrate that our method outperforms several existing state-of-the-art approaches.
https://arxiv.org/abs/2610.34618
Indoor context deciphered by ambient sensing underpins many applications, from assisted living to smart buildings, yet remains weakly connected to the language-aligned paradigm of modern multimodal systems. Through billions of IoT devices, ambient WiFi Channel State Information (CSI) captures the indoor context of human activity and environmental dynamics passively via ubiquitous, always-on infrastructure. However, unlike vision or audio, CSI lacks naturally paired text like captions or transcripts, limiting its integration with language-driven AI systems. We present CSI-TextBench, a large-scale dataset and benchmark for aligning CSI data with natural language, unifying 3 sensing benchmarks into 285,790 samples across 7 tasks and pairing them with 2,291 hierarchically structured, label-derived descriptions via many-to-many assignment to construct 2.0 million CSI-text training tuples. Benchmarking 5 alignment methods from 4 paradigm families, we show that contrastive CSI-text alignment matches supervised classifiers without class labels, adapts to unseen classes with few samples, and responds to open-vocabulary natural-language prompts. A single unified model supports compositional scene understanding, answering joint queries over activity, proximity, and identity, achieving 86.9% top-1 accuracy on 140 composite scenes, outperforming task-specific classifier ensembles. CSI-TextBench enables ubiquitous ambient sensing as a language-indexed perception layer, enabling indoor contextual awareness in language-driven AI systems.
https://arxiv.org/abs/2610.34616
Autonomous agents have been observed acting beyond their assigned tasks, accessing unauthorized systems, and establishing communication channels to exchange information, yet the Web and network-level techniques they use to extend their capabilities and circumvent security restrictions remain poorly understood. We study these techniques through the ProWiki incident, in which agents restricted to outbound HTTP GET requests repurposed a public German-language wiki as a message board. We analyze 14,591 agent edits containing 79,541 URL postings and present corroborating evidence that the published URLs reflect the agents' operational activity. We find that 26,717 URL occurrences (33.6%) embed target URLs within one to five layers of third-party Web services. By recursively reconstructing these nested URLs, we identify 67 distinct intermediate services and classify their functionality. Most intermediate layers (73.8%) extend agents' capabilities by delegating content extraction and transformation to external Web services: Markdown and OCR services convert web content into machine-readable text, JSON filtering services extract selected fields, and archival services expose historical versions of pages. The remaining 26.2% facilitate circumvention of access restrictions and evasion of detection, including CORS and fetch proxies that retrieve content on agents' behalf, and link shorteners and redirectors that hide destination URLs from mechanisms inspecting literal URLs or outer hostnames. Agents also reach the same targets through multiple URL paths, consistent with attempts to evade rate limits or obscure request origins. Our results expose a gap between the capabilities granted to agents and those they can obtain by composing publicly available Web services: restricting an agent's local toolset or direct Internet access may be insufficient while third-party services remain reachable.
https://arxiv.org/abs/2610.34542
Spatial audio authoring coordinates source decisions with soundfield synthesis. For editing, plans specify what to change and retain, while reference audio supplies further acoustic detail. We present AMBIT, a unified framework for native First-Order Ambisonics (FOA) generation and editing with three connected contributions. First, a shared Transfusion backbone couples autoregressive planning with rectified-flow rendering through executable ScenePlans that specify source content, activity, and 3D trajectories. For editing, the plan describes the full target scene, including retained sources; contrastive features ground the requested change, while full reference latents supply residual acoustic detail for preservation. Second, an FOA-aware continuous VAE combines shared-decoder omnidirectional recovery, asymmetric grouped KL, and spatial covariance matching to preserve content and inter-channel structure. Third, joint Transfusion on-policy self-distillation (OPSD) refines planning on student-generated prefixes and rendering at sampling states reached under student plans, with target-conditioned teachers supervising both modes of the shared student. Evaluations across speech, music, and sound show improved generation content and spatial quality, acoustic fidelity during instruction-guided editing, and unedited-source preservation over the compared baselines. Project page: this https URL
https://arxiv.org/abs/2610.34415
Indoor IoT cameras can support assisted living and activity-aware spaces, but exporting raw RGB frames raises privacy concerns by exposing facial, clothing, and household details that body-motion tasks often do not require. We present StickFigure, a pose-only IoT camera system that, during real-time monitoring, keeps raw frames local and ephemeral while exporting only compact skeleton streams sufficient for downstream human activity recognition (HAR). While pose estimation has matured on GPUs, extracting accurate pose streams on resource-constrained IoT cameras remains difficult. To overcome these challenges and enable on-device pose streams on (almost) any IoT camera, we present two core designs in StickFigure: an activation-aware pose backbone that allocates channel capacity after spatial downsampling and a lightweight hierarchical MLP refiner that selectively corrects elbow and wrist coordinates using intermediate visual features and kinematic context. We fully implement StickFigure and evaluate it on public benchmarks and real-world deployment. Compared with state-of-the-art MCU pose models, StickFigure achieves the highest pose accuracy and downstream HAR utility on both NTU RGB+D and Penn Action, while using 96.5% fewer parameters and 99.3% fewer MACs per frame than RTMPose-S-based baselines, achieving 0.107 s p95 latency on a representative STM32H753 MCU. We believe StickFigure establishes a new operating point for privacy-aware edge vision on massive IoT cameras.
https://arxiv.org/abs/2610.34130
The pervasive deployment of Deep Neural Networks (DNNs) across the computing spectrum has brought growing attention to energy efficiency, alongside raw performance. While voltage scaling is a potent knob for power reduction, its application in next-generation nodes is hampered by severe throughput degradation and susceptibility to process variation. This work aims to address these energy-reliability trade-offs by introducing a hardware-software co-design methodology, based on the observation that timing-robust bit ranges within the accumulator of a systolic array vary significantly across different voltage-frequency operating points. At the software level, we propose an Accumulation-Aware Training approach that incorporates this information as a first-order optimization constraint during training. By integrating mixed-precision quantization with unstructured pruning, we effectively constrain the model's weight distribution to avoid activating the hardware's sensitized high-rank carry chains. At the architectural level, we complement this with selective MSB-masking and fine-grained clock-gating to minimize unnecessary dynamic switching activity. Furthermore, we employ a layer-wise Dynamic Frequency Scaling strategy to further adapt to the accumulator's reduced dynamic range, based on the numerical sensitivity of each layer. We experimentally evaluate the proposed solution on a deep scaled 2nm technology node and demonstrate that, when operating at a low-voltage frontier, our co-optimized post-layout architecture delivers substantial energy and energy-delay product gains of 39% and 23% respectively, relative to the nominal voltage implementation, while maintaining accuracy degradation within 1%.
https://arxiv.org/abs/2610.33824
Continuous physiological monitoring is essential for linking behavior, internal state, and neural activity in neuroscience and translational research. In common marmosets, a key nonhuman primate model for social neuroscience, respiratory rate provides a basic readout of arousal, stress, and physiological regulation. Contact sensors can provide accurate measurements, but they add preparation burden and may interfere with head-free behavioral protocols. Thermal video offers a noncontact alternative by capturing breathing-related temperature changes around the nose and mouth. However, in head-free marmosets, these cues are weak, localized, and easily corrupted by head motion, pose changes, and fur. Here we show that this setting exposes a broader challenge in video-based physiological sensing: visually reliable tracking does not necessarily imply physiologically valid measurement. Using synchronized thermal video and contact respiration recordings, we show that marmoset respiration is recoverable from thermal video but strongly ROI-sensitive. Under head-free motion, visually plausible nasal tracking can still produce traces inconsistent with contact respiration, demonstrating that tracking is not sensing. We therefore estimate respiratory rate from learned temporal signals extracted from thermal clips and introduce Selective Temporal Attenuation Regularization (STAR). STAR selectively attenuates high-motion thermal discontinuities through local temporal imputation during training, regularizing models toward respiration-consistent dynamics while leaving inference unchanged on raw video. Across multiple deep-learning backbones, STAR improves head-free estimation, reaching 2.23 bpm MAE compared with 20.76 bpm for tracked ROI analysis.
https://arxiv.org/abs/2610.33661
Large language models (LLMs) are increasingly used as user simulators, but can they faithfully reproduce an individual's behavior and decision logic over time? We investigate this question in a longitudinal paper-trading study with 80 participants. Using rolling next-day predictions, we evaluate trading activity, buy-sell structure, asset selection, and portfolio outcomes, complemented by controlled input ablations and participant-rated explanation alignment. No evaluated LLM reliably outperforms a simple recent-activity persistence baseline for predicting trade occurrence, while buy-sell structure and asset selection remain difficult to reproduce. Similar activity-level performance can also yield substantially different portfolio trajectories. Ablations show that activity predictions depend primarily on recent trades, whereas asset choices are sensitive to multiple evidence sources. Participants rate explanations as more aligned with their recurring strategies than with their decision-specific rationales. These findings reveal a gap between coarse behavioral agreement and individual-level fidelity, motivating separate evaluation of behavioral accuracy, explanation alignment, and evidence dependence in personalized user simulation.
https://arxiv.org/abs/2610.26524
This paper proposes a five-level organization of message formation and understanding within thinking and activity in which ASI participates. Its central thesis is that a language model's context is fundamentally incapable of presenting the foundations of utterance formation for examination and verification. These foundations are established in advance by acquired schemas, systemic representations of the world of activity, and connections accepted as true within them. Neither context nor trained LLM weights, in themselves, attest to the presence of deliberately constructed systemic representations. The five levels distinguish systemic representations of the world of activity, a particular situation cut out from those representations, its narrative description, the full sense of a message, and text. A situation's schema-model is not identical to a story about it. A recipient reconstructs the full sense from an abbreviated text through jointly constructed and agreed-upon representations and joint activity. Reverse work proceeds from the full sense through an extended description and the situation's schema-model to the reconstruction of systemic foundations. The theoretical foundations draw on the systems methodology of G. P. Shchedrovitskiy, the structural differential of A. Korzybski, and the Meaning-Text theory of I. A. Mel'chuk. The contribution is a specific systemic-activity organization of content for examining and adjusting message foundations and controlling the consequences of their transformation. Further research concerns large graphical models of activity on reconstructible systemic foundations, distinct from text generation from context.
https://arxiv.org/abs/2610.32869
Surgical scene understanding is a prerequisite for context-aware and intelligent systems in the operating room. Current methods rely mostly on visual data, while acoustic sensing has shown potential to provide complementary information on surgical activity. First approaches for the localisation of surgical activity relied on microphone arrays and acoustic beamforming. This work investigates whether the active instrument can be localized from a single microphone and a camera by learning the correspondence between the appearance and the sound of surgical instruments, without annotating its position. We adapt weakly supervised audio-visual localisation, which learns from the co-occurrence of image and sound, to the surgical setting. We record simulated surgical actions, sawing, drilling, and chiseling, and train the model on these recordings without any spatial annotation to produce a heatmap over the sounding tool. We extend the training objective by two terms: a silence term, which suppresses the output for silence and background noise, and a composite term, which trains the model to select the sounding one of two visible instruments. The model localizes the active instrument on an independent corpus recorded with a different camera and microphone, in which instrument positions were obtained by optical tracking. It separates tool activity from silence and background noise and can attribute sound correctly when two tools are present. Audio-visual correspondence learned from in-domain recordings connects the sound of a surgical action to the instrument that produces it, using a single microphone and a camera. This proof of concept introduces a novel paradigm that enables the detection and localisation of surgical activity without manual annotations.
https://arxiv.org/abs/2610.32816
Human activity recognition benchmarks underrepresent culturally situated forms of movement and interaction. We introduce Ampe-DB, a multimodal dataset based on Ampe, a traditional Ghanaian game involving coordinated jumping, leg movements, and two-person interaction. It contains 637 canonical clips from 15 participants, with RGB video, two-person pose representations, participant and pair metadata, four-class movement labels, and derived binary movement-relation labels. We evaluate four-class movement recognition and binary movement-relation classification using three protocols: a duplicate-safe fixed benchmark, four-fold pair-held-out evaluation, and participant-held-out diagnostics for participants with complete class coverage. Baselines comprise a frozen Kinetics-400-pretrained R(2+1)D-18 RGB encoder, a two-layer bidirectional GRU for pose sequences, and validation-selected RGB-pose late fusion. For four-class recognition, RGB macro-F1 falls from 0.955 on the fixed benchmark to 0.418 under pair-held-out evaluation and 0.384 under participant-held-out evaluation. Pose is more robust, achieving 0.957, 0.811, and 0.839 across the same protocols. Fusion reaches 0.976 on the fixed benchmark and 0.824 under pair-held-out evaluation, but does not consistently improve on pose under participant shift. Binary movement-relation recognition also becomes substantially harder under held-out evaluation, especially for different-leg configurations. These findings show that strong fixed-benchmark scores can mask failures to generalize across people and pairings. Ampe-DB supports research on fine-grained movement recognition, human-human interaction, multimodal learning, and identity-aware generalization.
https://arxiv.org/abs/2610.33130
The 3rd WEAR Dataset Challenge asks for an activity label for every one-second window of four unseen participants. Each window has 15 VideoMAEv2 frame embeddings, the accelerometer signal of one random limb, and the windows are shuffled. This report is an empirical analysis of one system built after the challenge deadline. Window classifiers are stacked with features averaged over per-participant graphs of video neighbours and scored successor candidates, and a mean-field style decoder with a soft count constraint per activity labels all windows of a participant together. The setting is offline and transductive, and the count constraint relies on each activity being performed once. In a participant-wise simulation with participant-level bootstrap intervals, stacking added 0.051 macro F1 over stage-1 fusion and the count constraint 0.011 (95% interval 0.006 to 0.017). A quadratic seam distance ranks the true same-limb successor first among about 720 candidates in 83.9% of cases, but adding it to the link model changed macro F1 by -0.0001 (-0.0006 to 0.0005). With the decoder settings and the mixing weight chosen inside the folds and the mislabelled spans left out of training, the system scores 0.936 in simulation (0.940 for participants who performed each activity once); the submitted configuration scores 0.938 (0.943). Per-participant scores range from 0.80 to 0.97. These results come from a non-nested stacking protocol that uses unlabelled features transductively, so they are not a fully inductive estimate. An oracle that averages the final unary scores over the true segments reaches 0.973, and 79% of the remaining errors involve the null class. On the private leaderboard, which no choice used, the selected Kaggle submission scored 0.9231 (public 0.9254, with bounds partly tuned there), and a later one with bounds chosen in simulation scored 0.9247 (public 0.9253).
https://arxiv.org/abs/2610.32398
Multi-task deep learning has been developed to ensure explainability in computer-aided diagnosis (CAD) systems. These systems simultaneously predict interpretable characteristics of findings in auxiliary tasks to help radiologists assess the reliability of the diagnostic prediction. However, the CAD's opinion would not be referred to when the auxiliary task prediction deviates from the radiologist's expectation. To further leverage CAD's knowledge in the above situation, we proposed a multi-task gradient assist (MGA) method that brings interactivity to multi-task models without additional training. MGA allows radiologists to give feedback to the CAD and ask for additional diagnostic suggestions. Using two datasets of mammography and CT, we show that MGA successfully improves diagnostic accuracy upon feedback on various auxiliary tasks. Compared to an existing method using forward propagation only, MGA performs more robustly and can be combined to further improve accuracy. Our method paves a promising way for integrating artificial intelligence into image diagnosis.
https://arxiv.org/abs/2610.32396
Point-supervised Temporal Action Localization (PTAL) aims to localize action instances using only a single annotated timestamp for each instance. Such sparse supervision offers limited guidance for modeling the structural organization of an action instance. Existing PTAL methods primarily infer action regions from classification responses, leaving the structural organization of multiple action-related parts within each instance insufficiently modeled. To address this problem, we propose DeformStruct, an instance-level deformable structure modeling framework for PTAL. DeformStruct progressively constructs the structure of each action instance through three stages. First, Structural Anchor Mining discovers multiple anchors from annotated points and action-responsive regions, providing structural nodes grounded in the current instance. Second, Deformable Part Alignment associates latent structural parts with these anchors through deformation-aware matching, allowing their temporal supports to adapt to instance-dependent variations. Third, Relational Structural Organization explicitly models the relative temporal relations among the aligned parts, forming a relational representation that characterizes how different action-related parts are organized together. Extensive experiments on THUMOS14, ActivityNet-v1.3, GTEA, and BEOID demonstrate consistent improvements in point-supervised temporal localization and validate the effectiveness of deformable structural modeling.
https://arxiv.org/abs/2610.32321