Computer Science (arXiv)

We present a novel view on feature evolution in Vision Transformers (ViTs) by visualizing the training process over two dimensions -- network depth (layer) and training time (epochs). We employ Sparse Autoencoders (SAEs) to extract candidate sparse features from CLS-token representations and compare their activation profiles across epoch--layer pairs. This allows us to study feature-level dynamics that are not directly visible from representation-level similarity measures. Furthermore, we demonstrate how this framework of feature evolution allows us to describe feature migration, the change in the layer where a feature is most detectable during training. Our experiments show that migration is concentrated early in training, occurs more often toward earlier layers than toward deeper layers, and declines as feature organization stabilizes. We further find that deeper layers stabilize earlier and more strongly than shallow layers. The results show that our approach can be employed as a tool for understanding how ViTs learn and evolve.

Autonomous vehicles require robust perception and decision-making capabilities to operate in diverse and unseen scenarios. While reinforcement learning and rule-based methods can provide effective control and safety mechanisms, their performance may degrade in situations requiring contextual reasoning. Large Language Models (LLMs) have demonstrated strong capabilities in understanding multimodal information and generating contextual reasoning, however, their use for direct vehicle control can introduce latency and hallucination risks. To address these limitations, a hybrid framework is proposed. This system uses an orchestrator to coordinate PPO-trained reinforcement learning and PID control, with LLM common-sense reasoning applied throughout the framework. LLM reasoning is further employed iteratively to refine the RL reward function for dynamic driving environments. The proposed framework is evaluated in highly randomized CARLA scenarios under diverse environmental and traffic conditions. The results demonstrate the potential of integrating LLM-based reasoning with conventional autonomous driving methods while retaining structured control and safety mechanism.

Digital engineering uses many models from different perspectives, creating a connected set of digital artefacts across a product's life cycle. Designers seeking a holistic view must navigate numerous models and views, requiring domain-specific software, languages, and representations. This can lead to getting lost in scattered information and the cognitive burden of mentally integrating details across diagrams. To overcome these issues, we developed the virtual environment GraphXplore. GraphXplore enhances perceptual and conceptual integration by linking all relevant visual items from different perspectives into an interactive, layered 3D graph displayed in virtual reality, providing a holistic view of the system. We compared GraphXplore with a conventional on-screen setup using a PowerPoint slide deck with model screenshots viewed on a desktop PC. In an experiment with N=33 volunteers (mainly industrial product design postgraduates and professors), we conducted a baseline usability study focused on fundamental information retrieval tasks for model-based design comprehension, as identifying basic model elements is the fundamental prerequisite in design reviews. Our findings indicate that for simple retrieval tasks, the correctness of answers, completion time, recall score, and perceived confidence are comparable in both environments. However, the virtual environment demonstrated practical advantages, achieving a "good" average System Usability Scale (SUS) score of 73.1 compared to the slide setup's borderline score of 66.4. Furthermore, GraphXplore users reported a lower perceived cognitive workload (mean NASA-TLX score of 43.8) compared to the traditional setup (50.4). Future work will enhance this experiment to yield empirical results across massive industrial datasets, refine GraphXplore, and extend to new design review objectives.

Video Temporal Grounding (VTG) faces significant challenges when natural language queries must distinguish between multiple events involving visually similar entities, particularly when relying on fine-grained visual attributes that are difficult to describe accurately in words alone. To address this, we introduce Image-Disambiguated Video Temporal Grounding (ID-VTG), a task that leverages multimodal queries combining a reference image and a text description to precisely localize segments where a specific instance performs a described action. To facilitate research, we construct two benchmarks: IDVTG-Gym, focusing on fine-grained, compositionally ordered gymnastics actions with athletes in similar uniforms; and IDVTG-InternVid, an open-world dataset featuring diverse entities (e.g., humans, animals, fictional characters) and significant temporal distractors. Methodologically, we propose the Visually-Guided Disambiguation Aggregation (VGD-Agg) framework based on a dual-branch fast-slow architecture. The fast branch efficiently generates preliminary event proposals, while the slow branch performs fine-grained frame-level matching between video frames and the reference image. We enhance discriminability via two learnable tokens: a Compare Token, which represents hard negatives to probe for the presence of the target instance (as referred to by the query image), and a Depress Value, which represents text-irrelevant events. Proposals that the Compare Token identifies as lacking the target instance are pushed toward the Depress Value, thus easing disambiguation via the text query. Extensive experiments validate our approach, which achieves state-of-the-art results on the proposed benchmarks. Code is available at https://github.com/oceanflowlab/ID-VTG.

Reliable multi-UAV communication control depends on predicting which aerial links will serve traffic before measurements are available. Radio world models (radio WMs) make such planning tractable, but their errors are nonuniform: a globally accurate model may still fail along high-demand corridors or association boundaries where rate errors reverse control decisions. This mismatch creates a learning challenge. Link queries must reduce decision-relevant channel uncertainty, while counterfactual trials must be filtered so that biased rollouts do not corrupt the policy. Existing acquisition and model-based control treat these budgets separately, valuing uncertainty, coverage, or optimistic return rather than risk reduction. We present RMWorld, a task-aware radio-WM framework that couples value-of-information channel calibration with credibility-diversity multi-trial selection. A biased propagation formula is corrected by a Bayesian residual, and each link is valued by its exact one-label reduction in locally linearized task-integrated posterior rate variance. Counterfactual branches are selected by a task-gated log-determinant objective, followed by conflict projection and fixed-batch validation. We derive the variance-reduction identity, prove posterior task-risk equivalence and the submodular greedy guarantee, and establish a scoped first-order non-interference result. Across 100 paired 3GPP trials RMWorld reaches 0.949~bit/s/Hz task-weighted RMSE, and across 30 severe-load DeepMIMO trials it reduces median backlog by 0.967 versus Ensemble UCB at 37.5\% more offline rollouts.

We study inference-time control for text generation in discrete diffusion language models, where the goal is to steer sampling toward sequence-level rewards without retraining. Prior work in this domain has focused on particle-based methods such as best-of-$n$ sampling and bootstrap sequential Monte Carlo, which may suffer from overoptimism and weight degeneracy, respectively. We address these limitations using \emph{nested} sequential Monte Carlo methods. We formulate nested SMC (NSMC) and fully-adapted nested SMC (FA-NSMC) for Feynman--Kac steering, identifying and correcting errors in prior formulations that lead to biased final estimates. We evaluate these methods on toxicity and fluency steering tasks, showing that NSMC and FA-NSMC consistently outperform best-of-$n$ and bootstrap SMC.

Large multimodal models (LMMs) have demonstrated strong OCR recognition capabilities, yet remain vulnerable to adversarial visual text that is readable to humans but challenging for models to localize and recognize. Existing OCR benchmarks mainly focus on natural or document-style text, while adversarial OCR evaluations remain limited in scale, task coverage, or region-aware evaluation. In this paper, we formulate adversarial OCR as a \textbf{grounded OCR perception} task and introduce \textbf{AdvSpot}, the first benchmark for grounded adversarial OCR evaluation. AdvSpot comprises 390 images with region-level annotations, spanning 5 primary categories and 13 fine-grained adversarial OCR types. To address this challenge, we propose \textbf{ArmorOCR}, a two-stage training framework for robust adversarial OCR perception. ArmorOCR first acquires missing adversarial OCR perception from privileged transformed observations through On-Policy Self-Distillation (OPSD), and then refines grounded OCR perception through Group Relative Policy Optimization (GRPO) with task-conditioned rewards for localization, recognition, full spotting, and visual question answering (VQA). Experiments on our AdvSpot, other adversarial OCR benchmarks, and general OCR benchmarks demonstrate that ArmorOCR consistently improves adversarial OCR perception while preserving competitive general OCR capability.

The detection of rare disease-associated cell subsets from high-dimensional single-cell measurements is critical for understanding diseases such as leukaemia and viral infections. CellCnn, a convolutional neural network (CNN) designed for this task, has demonstrated the ability to identify phenotype-associated cell populations at frequencies as low as 0.01\%. Training such models reliably requires patient cohorts that are larger and more diverse than any single institution can typically assemble, and the underlying single-cell data is too sensitive to share across institutional boundaries under existing privacy regulations. We propose a secure multi-party computation (MPC) framework that enables the training and inference of CellCnn entirely on secret-shared data. This ensures that neither the participants nor the computing servers ever observe raw patient data or intermediate values. Evaluated on benchmark single-cell datasets for cytomegalovirus infection (CMV) and acute myeloid leukaemia (AML), our implementation preserves accuracy close to its plaintext counterpart while outperforming the prior privacy-preserving baseline. In contrast to earlier privacy-preserving approaches that removed components such as ReLU activations and bias terms, our method retains these key parts of the CellCnn architecture and supports accurate analysis without exposing raw patient data.

The emergence of large-scale Weather and Climate (W&C) datasets offers new opportunities for modeling extreme Earth events (ExEE) and their impacts using deep learning. However, their adoption in operational settings remains limited by the lack of models' interpretability. While for conventional text and image modalities, tools such as Sparse Autoencoders (SAEs) have proven effective for extracting human-understandable concepts, their use for the analysis of ExEE remains challenging due to the nature of W&C data. To address this, we introduce (i) a geographic location-based modulation of the inputs of SAE to capture the local semantic meaning of environmental patterns, and (ii) an ensemble of rule-based SAE-Xplainers to interpret the resulting high-dimensional features derived from complex, multi-modal environmental predictors. We evaluate our method on three ExEE types: the prediction of fires, and the detection of tropical cyclones and atmospheric rivers. We show that SAE input modulation improves both reconstruction performance and feature utilization, and that our SAE-Xplainers enable faithful interpretation of complex climatic patterns by unfolding them into human-understandable rules that are consistent with the scientific literature, while also supporting the identification of feature absorption.

Large language models (LLMs) are increasingly used in settings where textual summaries, numerical observations, and external tool outputs may provide conflicting evidence. We study how LLMs arbitrate between such sources when they support opposing decisions. To do so, we introduce a controlled synthetic benchmark in which latent risk trajectories generate both numerical time series and natural language summaries, allowing us to construct conflicts where exactly one evidence source is aligned with the ground-truth label. This design lets us independently manipulate modality, temporal recency, source reliability, and evidence provenance. Across open-weight instruction-tuned models, we find that arbitration behaviour is systematic rather than random: models exhibit distinct text-versus-number preferences, follow temporal recency more consistently than explicit reliability cues, and can over-rely on external forecasts even when they conflict with direct contextual evidence. These results suggest that current LLMs often rely on heuristic arbitration strategies when integrating heterogeneous evidence, highlighting a failure mode for tool-augmented decision systems.

Mobile manipulation requires a robot to predict how locomotion and arm motion jointly alter future observations and control. Existing world-action models, developed largely for fixed-base platforms, do not explicitly distinguish camera ego-motion from base and arm actions. Here we introduce DECOWAM, a whole-body world-action model that separates these factors through dedicated conditional interfaces. DECOWAM freezes an adapted FastWAM backbone and trains residual adapters, an action-equivalent future bottleneck distilled from privileged observations, adversarially separated base and arm latents, and base-velocity conditioning for video prediction. We further introduce ARMDOG, a real-robot dataset that synchronizes video, whole-body state and action, and language. On a fixed replay protocol, DECOWAM improved both future-video and action prediction over FastWAM, reducing action MSE by 21.7% with 25.95M trainable adaptation parameters. Across 79 closed-loop trials per method, it achieved the highest observed whole-body coordination and base-displacement robustness among the compared systems, while task completion remained comparable to the strongest baseline. These results show that embodiment-aware factorization can support parameter-efficient joint visual prediction and whole-body control under moving viewpoints.

Generative models have shown strong potential for positron emission tomography (PET) image reconstruction. Although diffusion model-based reconstruction methods have demonstrated promising performance, they often require many reverse sampling steps with data-consistency updates incorporated into the sampling process. Flow matching offers an attractive alternative because it can directly estimate clean images from intermediate states, allowing data-consistency refinement to be separated from flow propagation. In this work, we proposed flow matching-based PET image reconstruction methods. We first established PET-FlowDPS by incorporating Poisson likelihood guidance with an expectation-maximization (EM)-based preconditioner into the FlowDPS framework. We then proposed a model-based PET reconstruction method that used a pretrained flow matching model as a prior, in which the flow-based prior, PET data refinement, and stochastic propagation were interpreted within an approximate Bayesian framework. Experimental results using [$^{\text{18}}\text{F}$]FDG brain PET datasets showed that the proposed method achieved better bias-variance trade-offs across different dose levels compared with other reference methods. These results demonstrated the potential of flow matching as a generative prior for quantitative PET image reconstruction.

End-to-end autonomous driving has evolved from camera-to-control regression toward planning-oriented systems that use structured representations, trajectory-level outputs, and increasingly realistic evaluation protocols. This survey reviews this transition across behavior cloning, conditional imitation learning, privileged distillation, BEV and vectorized planning, unified perception-prediction-planning architectures, world-model-based planners, and vision-language-action systems. We argue that the key distinction in modern end-to-end driving is not whether intermediate representations are used, but whether they are learned, supervised, and evaluated to support safe, feasible, and route-compliant planning. To organize the literature, we synthesize existing methods along four axes: input representation, planning output, supervision signal, and evaluation protocol. We further examine the benchmark shift from open-loop trajectory matching to closed-loop simulation, non-reactive real-log evaluation, long-tail testing, and human-preference-aware metrics. Our analysis highlights that architectural progress is difficult to interpret without benchmark-consistent evaluation, and that displacement-based open-loop metrics alone provide limited evidence for safe and human-aligned driving. We conclude with open challenges in uncertainty-aware planning, learner-expert mismatch, runtime safety assurance, language-action grounding, world-model validation, and reproducible benchmarking.

Oscillatory Neural Networks (ONNs) are an alternative computing paradigm for AI and combinatorial optimization problems. However, digital architectures are often designed for specific applications of ONNs. This work introduces a modular and scalable architecture called ODEONN that is generic to multiple applications of ONNs, and to the best of our knowledge, is the first fully digital ONN to also support complex-valued coupling. Additionally, an approximation of the sine function is introduced that uses half of the hardware resources compared to standard methods. The performance of ODEONN is compared with a full-precision software simulation, where a performance degradation of less than $2\%$ is shown. Therefore, we conclude that the fixed-point quantization and the approximated waveform affect the accuracy of computation by only a small amount. Furthermore, ODEONN shows a 45$\times$ reduction in energy-delay product over the software simulation running on conventional hardware.

The field of usable security and privacy research is a young and expanding field, which is still developing standards for its research, e.g. regarding replications. We used a mixed-method approach, in order to get a better understanding of the current state of replications in the field of usable security and privacy: (1) we examine the Call for Papers of 13 venues spanning security, privacy, and human-computer interaction; (2) we conduct a systematic search for papers reporting replicated user studies published across these venues between 2016 and 2025, yielding 24 relevant publications; (3) we categorized these 24 papers employing the replication taxonomy proposed by Olszewski et al. (2025); (4) we distributed a survey to the authors of these papers to understand their motivations for conducting replications. Our analysis reveals four key insights: (A) Calls for Papers would benefit from clearer guidelines for authors and reviewers regarding replication work; (B) determining what modifications were made relative to the original study proves difficult when reading replication papers; (C) strict exact replications do not exist in our sample. Approximately two-thirds of the 24 studies altered multiple aspects of the original work; (D) temporal and contextual changes affecting results emerged as one of the most frequently cited motivations for replication. Based on these findings, we offer practical recommendations for venues, researchers, and peer reviewers to strengthen replication practices in usable security and privacy research.

Recent advances in generative video models have significantly improved visual realism in video object removal, yet evaluation protocols still focus on masked region fidelity, treating removal as local inpainting. In real scenes, object removal is a causal intervention: eliminating an object also requires removing its induced physical effects, such as shadows, reflections, illumination changes, translucency, and dynamic traces. Existing benchmarks lack aligned clean references or remain limited to simplified synthetic settings, preventing systematic evaluation of causal consistency. We introduce BeyondMasks, a paired benchmark for causally consistent video object removal, consisting of temporally aligned synthetic and real world video pairs with clean background references. The dataset spans diverse photometric, geometric, volumetric, and dynamic interactions, and supports both mask based and instruction driven editing. We further propose CORE, a structured vision language model based evaluation protocol that jointly measures object disappearance and after effect consistency, aligning more closely with human judgments than existing metrics. Benchmarking state of the art methods reveals systematic failures in removing secondary physical effects despite high masked region fidelity, exposing a gap between visual plausibility and causal correctness. BeyondMasks reframes video object removal as causal scene consistency rather than local reconstruction and provides a unified framework for its evaluation.

We introduce OenoBench, a wine-domain knowledge benchmark of 3,266 multiple-choice questions across six pillars (regions, grape varieties, viticulture, winemaking, producers, business) and four difficulty tiers. The corpus is built from 38,104 atomic, source-anchored facts extracted by 35 provenance-verified scrapers from government registries (INAO, TTB, OIV), peer-reviewed journals, and Wikipedia/Wikidata. Our methodological contribution is an LLM-driven pipeline in which language models reformat verified facts and audit the result, but never serve as the source of truth: every claim traces to a URL, every question is generated by one of five strategies across five generator families, and every question is scored by a nine-agent audit calibrated against a human gold sheet via Cohen's $κ$. Evaluating sixteen frontier configurations, we find: (i) overall accuracy spans 53%-84%, led by o3 at 83.6%; (ii) reasoning-mode lift concentrates in DeepSeek R1 (+6.8pp) and is absent in Claude Opus and Gemini Pro; (iii) Anthropic shows +9pp self preference on its own questions while Google shows -8pp inverse preference; (iv) frontier open-weight models share the cost-vs-accuracy Pareto frontier with proprietary reasoning models; and (v) every config gains around 33pp on closed-book solvable items, revealing a parametric-recall ceiling that only the contextual slice avoids. We release corpus, audit findings, human-review app, and construction code under CC-BY-SA-4.0.

Artificial immune networks (AINs) are naturally memory-forming systems, but conventional visual AINs often rely on flattened vector affinity that ignores spatial structure. This paper studies whether structured, gradient-free immune affinity can make Deep AINs viable as replay-free visual class-incremental representation-memory learners. Visual B-cells are formalized as structured templates, including shifted-template affinity, zero-normalized cross-correlation (ZNCC) filters, and feature-map binding profiles. A repertoire is treated both as memory and as a representation-inducing basis, while depth is obtained by passing binding-profile response maps to subsequent immune layers. The resulting Deep AIN exhibits adaptive latent coordinate reorganization: as new classes arrive, the binding-profile space evolves while retaining recoverable structure for earlier classes. Experiments on sklearn digits, MNIST, Fashion-MNIST, and KMNIST show that preserving response maps is critical. Scalar binding-profile variants underperform, whereas feature-map Deep AINs learn class-discriminative visual memory without replay, label-driven immune updates, or backpropagation through the immune layers. On sklearn digits, downstream probes fitted on the learned binding profiles reach 0.939 final balanced accuracy with logistic regression and 0.902 with 1-nearest-neighbour after all ten classes are encountered, with initial-class retention of 0.978. Adaptive layer-wise scale calibration further improves the two-layer feature-map Deep AIN to 0.978 balanced accuracy. With the same calibration rule, Fashion-MNIST reaches 0.814 and KMNIST reaches 0.853. These probes are external validation tools, not components of the AIN. The results identify structured affinity, response-map preservation, adaptive latent reorganization, and layer-wise scale calibration as key mechanisms for replay-free visual immune memory.

LLM-based Multi-Agent Systems (MAS) achieve strong performance on complex reasoning tasks by coordinating multiple agents, but at the cost of substantial token consumption. Recent work on automatic topology design, ARG-Designer, has reframed this problem as autoregressive graph generation. However, its training objective provides no explicit incentive for the model to generate sparse and efficient topologies. We address this limitation by introducing a Reward-Guided Autoregressive Graph Generation (RGA-Designer) inspired by Reinforcement Learning from Human Feedback (RLHF). We train a reward model that jointly captures task correctness and structural compactness, and then fine-tune the pretrained graph generator using the reward model as feedback. Our method preserves task accuracy at the level of ARG-Designer while reducing token consumption by an average of 20.5%.

Retrieval-Augmented Generation (RAG) lets Large Language Models (LLMs) pull in up-to-date, domain-specific information instead of relying only on what they were trained on. Yet most RAG systems still draw from centralized databases with limited oversight, making it difficult to verify where a document came from, whether it has been tampered with, or whether it should be trusted at all. This is a serious problem in domains where both the timeliness and accuracy of retrieved content are critical, such as healthcare, finance, logistics, and legal case law, where a wrong or manipulated document can directly lead to bad decisions. We present TrustRAG, a committee-based, blockchain-backed RAG system: before a document is used, it is certified by a committee of domain experts through a zero-knowledge protocol, and the committee's hidden scores are combined via secure multi-party computation into a trust score that any client can verify. These scores, along with the underlying document data, are maintained jointly across chains through hash commitments, so no document or score can be silently altered or dropped, and every ranking can be independently replayed and checked.

In this work, we present HandMvNet, one of the first real-time method designed to estimate 3D hand motion and shape from multi-view camera images. Unlike previous monocular approaches, which suffer from scale-depth ambiguities, our method ensures consistent and accurate absolute hand poses and shapes. This is achieved through a multi-view attention-fusion mechanism that effectively integrates features from multiple viewpoints. In contrast to previous multi-view methods, our approach eliminates the need for camera parameters as input to learn 3D geometry. HandMvNet also achieves a substantial reduction in inference time while delivering competitive results compared to the state-of-the-art methods, making it suitable for real-time applications. Evaluated on publicly available datasets, HandMvNet qualitatively and quantitatively outperforms previous methods under identical settings. Code is available at github.com/pyxploiter/handmvnet.

Ontology-based data access is intended for graph data, but practical support remains limited to SQL-like languages lacking the navigational and path-matching features fundamental to graph querying. Theoretical algorithms for navigational queries have long been available. Still, they have never been implemented, largely because practical graph query languages fell short of their theoretical counterparts and lacked the expressive power needed to support rewriting common ontology languages. The recent standardisation efforts around GQL and SQL/PGQ finally allow us to overcome this barrier and present a practical query rewriting technique for ontology-mediated navigational graph queries. Ontologies are written in a DL-Lite variant tuned to property graphs, and the queries in a GQL fragment with nested two-way regular path queries and which can be evaluated in Cypher. Preliminary experiments with our proof-of-concept prototype suggest that querying graph data with ontological knowledge may finally be within reach.

The term "visualization design" encompasses multiple concepts and skills that go well beyond current assessments of graphical perception and visualization literacy. In the context of education, what exactly should a student be able to do if they "know" visualization design? To answer this question, we draw on existing methodology from the field of education to propose a concept inventory for visualization design, i.e., a theoretical model capturing the most important concepts and skills commonly associated with visualization design. We initially draft the concept inventory using a qualitative analysis of course objectives from visualization course syllabi. Then, we iteratively refine the concept inventory by soliciting feedback from instructors through semi-structured interviews. Based on our experiences in developing the concept inventory, we reflect on open questions and future research directions in visualization education, such as developing assessments for visualization design (similar to those for visualization literacy) and providing automated assistance for learning and teaching visualization design. Our supplemental materials are available at https://osf.io/dma3f/.

Humanoid robots have recently demonstrated promising capabilities in real-world ball sports. However, achieving professional motion styles while maintaining strong task performance remains challenging. In this work, we propose AdaPT, an Adaptive Motion Planning and Tracking framework that learns professional tennis serving and rally styles directly from broadcast videos. This hierarchical design is motivated by the key insight that the planner generates stylistic kinematic motions, while the tracker executes them with minimal interference with planning. Despite its effectiveness in simulation, a substantial sim-to-real gap emerges: tracking performance inevitably degrades on real robots, and this degradation is partially overlooked by autoregressive planning and further compounded by noisy perception. To address these issues, our adaptation mechanism improves tracking robustness by learning to track randomized execution speeds, while conditioning the planner on a learned motion-speed adapter to mitigate compounding errors. Real-world experiments on the Unitree G1 demonstrate the effectiveness of our adaptation mechanism in bridging the sim-to-real gap. We further deploy AdaPT policies on the full-size Dobot Atom humanoid robot (1.7m) and demonstrate in-the-wild serving without motion capture. Beyond these results, our real-world experiments reveal both algorithmic and engineering insights for future humanoid ball-sports systems. Videos and code are available on our \href{https://humanoidtennis.github.io/AdaPT/}{project website}.

Robots executing long-horizon manipulation tasks from natural-language instructions must reason about both semantic task structure and geometric feasibility. However, under partial observability, the availability of goal-relevant objects may be uncertain. In such cases, approaches that combine Vision-Language Models (VLMs) with Task and Motion Planning (TAMP) may generate subgoals that rely on the VLM's prior knowledge without observational support, leading to execution failures or unintended outcomes. We propose Evidence Acquisition and Feasibility Gating (EAFG), a framework that acquires visual evidence through VLM-generated exploratory subgoals and TAMP-based execution. EAFG then applies a feasibility gate to decide whether to proceed with task planning, acquire further evidence, or halt. Our experiments show that, in cooking tasks with ambiguous object use, EAFG improves recipe completion by discovering task-relevant objects before planning. For instructions requiring an absent object, EAFG promotes appropriate halt decisions and reduces repeated attempts to manipulate that object.