# object Published articles for object. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Align Your Query: Representation Alignment for Multimodality Medical Object Detection DevFeed: [Align Your Query: Representation Alignment for Multimodality Medical Object Detection]() Original publisher: [Read original article]() Author: Ara Seo, Bryan Sangwoo Kim, Hyungjin Chung, Jong Chul Ye Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Medical imaging](), [Representation learning](), [Embeddings]() Tags: [abstract](), [align](), [alignment](), [annotations](), [approach](), [architectures](), [arxiv](), [attention](), [challenge](), [class](), [context](), [cs-ai](), [cs-cv](), [cs-lg](), [define](), [detection](), [effective](), [effectively](), [medical](), [multimodality](), [object](), [object-detection](), [query]() ### AI overview For medical object detection trained across modalities such as CXR, CT, and MRI, the paper proposes aligning DETR-style object queries with modality context to address heterogeneous data statistics and disjoint representations. MoCA uses self-attention to propagate compact, text-derived modality tokens through the query set; QueryREPA adds a short pretraining stage with a modality-balanced contrastive objective to align queries with those tokens. The abstract reports consistent AP improvements across jointly trained modalities, with minimal latency and no architectural modifications. It does not provide numerical results or further evaluation details. ### Source excerpt arXiv:2510.02789v3 Announce Type: replace-cross Abstract: Medical object detection suffers when a single detector is trained on mixed medical modalities (e.g., CXR, CT, MRI) due to heterogeneous statistics and disjoint representation spaces. To address this challenge, we turn to representation alignment, an approach that has proven effective for bringing features from different sources into a shared space. Specifically, we target the representations of DETR-style object queries and propose a simple, detector-agnostic framework to align them with modality context. First, we define modality tokens: compact, text-derived embeddings encoding imaging modality that are lightweight and require no extra annotations. We integrate the modality tokens into the detection process via Multimodality Context Attention (MoCA), mixing object-query representations via self-attention to propagate modality context within the query set. This preserves DETR-style architectures and adds negligible latency while injecting modality cues into object queries. We further introduce QueryREPA, a short pretraining stage that aligns query representations to their modality tokens using a task-specific contrastive objective with modality-balanced batches. Together, MoCA and QueryREPA produce modality-aware, class-faithful queries that transfer effectively to downstream training. Across diverse modalities trained altogether, the proposed approach consistently improves AP with minimal overhead and no architectural modifications, offering a practical path toward robust multimodality medical object detection. ## HaNDF: Object-Conditioned Geometric Neural Hand Distance Fields DevFeed: [HaNDF: Object-Conditioned Geometric Neural Hand Distance Fields]() Original publisher: [Read original article]() Author: Zuhao Liu, Melih Darcan, Zhengdi Yu, Riza Alp Guler, Tolga Birdal Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [signed distance functions](), [Artificial Intelligence](), [Neural Network](), [3D]() Tags: [computer-vision](), [cs-ai](), [cs-cv](), [cs-lg](), [cs-ro](), [data-driven](), [fields](), [generative](), [interaction](), [manipulation](), [neural](), [object](), [plausible](), [synthesis]() ### AI overview HaNDF models plausible hand poses in relation to an interacting object using an object-conditioned neural distance field. Plausible articulations lie on the field's zero-level set; Riemannian geometry tools are used to represent that set and project a given pose onto it. The method can also optimize for a plausible conditioning object. The abstract identifies pose refinement, reconstruction under occlusion, and physically plausible manipulation synthesis as potential applications, but provides no implementation details or specific evaluation results. ### Source excerpt arXiv:2611.08578v1 Announce Type: cross Abstract: We propose, HaNDF, object-conditional neural distance fields as data-driven priors for modeling the plausible hand-object configurations, in the product manifold of hand articulations. While recent unconditional generative priors showed success in modeling human pose, hands are versatile articulated bodies often found in interaction with other objects. This makes it crucial to develop a generative prior which can model the plausible hand poses conditioned on the object under interaction. To this end, our HaNDF proposes a geometry-aware, object-conditioned neural distance field, representing the plausible articulations in the zero-level set of a conditional field. We leverage tools from Riemannian geometry to (i) represent plausible hands in the zero level-set of an object-induced neural field, and (ii) project a given pose onto this field. Using HaNDF, we can also optimize for a plausible condition, determining the object under interaction. Our extensive evaluations demonstrate that our HaNDF provides a flexible and powerful prior for hand-object interaction, suitable for applications such as pose refinement, reconstruction under occlusion, and physically plausible manipulation synthesis. ## Panoptic Scene Program Diffusion Transformer DevFeed: [Panoptic Scene Program Diffusion Transformer]() Original publisher: [Read original article]() Author: Chika Maduabuchi Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [diffusion-transformers](), [text-to-image](), [Image]() Tags: [architecture](), [consistency](), [counts](), [cs-cv](), [cs-lg](), [diffusion](), [grounding](), [identity](), [image](), [inference](), [models](), [object](), [ordering](), [overhead](), [ownership](), [program](), [psp](), [quality](), [robust](), [settings](), [spatial](), [transformer]() ### AI overview PSP-DiT targets compositional prompt errors by jointly denoising image latents and panoptic scene-program latents in coupled transformer streams. Panoptic grounding and cycle-consistency objectives connect instances, attributes, relations, and counts to visual evidence in the generated image. Under matched training and inference settings, it outperforms a strong flat-text baseline on GenEval 2, SANEval-Simple, PSG-Score, and DetailMaster, with the largest gains reported for counting, attribute binding, role-sensitive relations, and long structured prompts. The abstract says image quality is preserved, inference overhead is modest, and performance remains robust to imperfect scene programs. ### Source excerpt arXiv:2609.31780v2 Announce Type: replace-cross Abstract: Modern text-to-image models produce high-fidelity images but still struggle with compositional prompts that require instance identity, attribute ownership, counting, spatial ordering, and role-sensitive relations. We introduce Panoptic Scene Program Diffusion Transformer (PSP-DiT), a diffusion-transformer architecture that treats a panoptic scene program as a first-class latent variable rather than an external control signal or post-hoc parse. PSP-DiT jointly denoises image latents and scene-program latents through coupled transformer streams, while panoptic grounding and cycle-consistency objectives tie object instances, attributes, relations, and counts to visual support in the generated image. Under matched training and inference settings, PSP-DiT improves over a strong flat-text baseline across GenEval 2, SANEval-Simple, PSG-Score, and DetailMaster, with the largest gains on counting, attribute binding, role-sensitive relations, and long structured prompts. The method preserves image quality, adds modest inference overhead, and remains robust to imperfect scene programs. ## CVE-2026-17405 - Multiple Vulnerabilities in IBM API Connect DevFeed: [CVE-2026-17405 - Multiple Vulnerabilities in IBM API Connect]() Original publisher: [Read original article]() Author: CVEFEED.io; us.ibm.com Published: 2026-11-07T05:18:57Z Content type: article Language: en Sources: [Latest High/Critical Vulnerabilitiy Feed]() Topics: [Authorization](), [Access Control](), [ibm]() Tags: [2026](), [api](), [api-connect](), [authorization](), [cve](), [cve-2026-17405](), [ibm](), [information](), [keystore](), [multiple](), [object](), [remote](), [user](), [vulnerabilities]() ### AI overview IBM API Connect versions 10.0.8.0-10.0.8.12 and 12.1.0.0-12.1.1.2 have a broken object-level authorization flaw in keystore access checks. Because the checks do not verify that a requester may read a keystore object, a remote authenticated attacker could obtain sensitive keystore information. The supplied excerpt gives a severity score of 8.8 (high) but no remediation steps. ### Source excerpt CVE ID :CVE-2026-17405 Published : Nov. 7, 2026, 5:18 a.m. | 32 minutes ago Description :IBM API Connect 10.0.8.0 through 10.0.8.12 and 12.1.0.0 through 12.1.1.2 could allow a remote authenticated attacker to obtain sensitive information due to broken object-level authorization. Access checks on keystore objects do not confirm that the requesting user is permitted to read them, which could expose keystore information. Severity: 8.8 | HIGH Visit the link for more details, such as CVSS details, affected products, timeline, and more... ## Executable but Unfaithful: Mock Object, Stub, and Assertion Generation by LLMs and Coding Agents DevFeed: [Executable but Unfaithful: Mock Object, Stub, and Assertion Generation by LLMs and Coding Agents]() Original publisher: [Read original article]() Author: Hao Ning, Jiahui Dong, Zhen Yang, Guanda Yan, Yichen Sun, Xuanyu Zhu, Zheng Shen, Yifei Pei, Jacky Keung, Jie M. Zhang Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Software Engineering]() Topics: [Mocking](), [AI-assisted coding]() Tags: [benchmark](), [coding-agents](), [cs-se](), [dependencies](), [detection](), [generation](), [java](), [language-models](), [llms](), [mock](), [mocking](), [object](), [source-code](), [study](), [tasks](), [test](), [testing]() ### AI overview MockBench evaluates generation of mock objects, stubs, and mock assertions across 165 tasks from real-world Java and Python projects, using six LLMs and three coding agents. It measures executability, behavior fidelity, and interaction-fault detection. Coding agents substantially improve executability over LLMs, but that advantage does not consistently yield more faithful behavior or better fault detection. Analysis of 660 failures identifies ten error types and shows how generated artifacts can pass tests while still deviating from intended behavior. The abstract contrasts these approaches with capture-and-replay methods, which require recorded dependency behavior, and StubCoder, which uses predefined symbol pools. ### Source excerpt arXiv:2610.14629v2 Announce Type: replace Abstract: Mocking is a foundational practice in unit testing, enabling developers to isolate the code under test by replacing real dependencies with simulated alternatives. Automating the generation of mock objects, stubs, and mock assertions has been a long-standing research goal. However, capture-and-replay approaches rely on recording the runtime behavior of actual dependencies. Evolutionary approaches such as StubCoder overcome this requirement but constrain generation to predefined symbol pools. Large language models (LLMs) and coding agents offer an alternative by generating mocking artifacts from source code without requiring recorded dependency executions or fixed symbol pools. Yet, no systematic study has evaluated their capabilities across all three artifact types. To address this gap, we construct MockBench, a benchmark comprising 165 mock artifact generation tasks from real-world Java and Python projects, with validated reference mock objects, stubs, and mock assertions. We evaluate six LLMs and three coding agents in terms of executability, behavior fidelity, and interaction-fault detection capability, complemented by comparisons with existing tools and sensitivity analysis. Experimental results show that LLMs and coding agents indeed have been the state-of-the-art techniques in mock-based testing. More specifically, compared with LLMs, coding agents substantially improve executability, while these gains do not consistently translate into higher behavior fidelity or better interaction-fault detection. Furthermore, we analyze 660 failures from LLMs and coding agents, and identify ten error types across the three artifact categories, revealing behavioral deviations that passing tests can conceal. Our findings highlight the challenges of mock artifact generation and provide practical guidance for both developers and researchers. ## Characterizing universal object representations across vision models DevFeed: [Characterizing universal object representations across vision models]() Original publisher: [Read original article]() Author: Florian P. Mahner, Johannes Roth, Ka Chun Lam, Michael F. Bonner, Francisco Pereira, Martin N. Hebart Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Artificial Intelligence]() Tags: [architecture](), [convergence](), [cs-cv](), [cs-lg](), [deep-neural-networks](), [dimension](), [dimensions](), [interpretability](), [it](), [models](), [object](), [performance](), [q-bio-nc](), [semantic](), [similarity](), [vision]() ### AI overview Across 162 vision models, the study decomposes object similarity structure into non-negative dimensions and measures how often each dimension recurs. Recurring, universal dimensions are more interpretable and more strongly tied to conceptual image properties than model-specific ones. Differences in architecture, objective, training data, model size, and performance do not explain which universal dimensions emerge. Models with more such dimensions better predict macaque IT activity and human similarity judgments, suggesting a link between representational universality and biological vision. ### Source excerpt arXiv:2605.13675v2 Announce Type: replace-cross Abstract: Deep neural networks trained with different architectures, objectives, and datasets have been reported to converge on similar visual representations. However, what remains unknown is which visual properties models actually converge on and which factors may underlie this convergence. To address this, we decompose the object similarity structure of 162 diverse vision models into a small set of non-negative dimensions. To determine universal versus model-specific dimensions, we then estimate how often each dimension reappears across models. In contrast to model-specific dimensions, universal dimensions are more interpretable and more strongly driven by conceptual image properties, indicating the relevance of interpretability and semantic content as implicit factors driving universality across models. Differences in architecture, objective function, training data, model size, and model performance do not explain the emergence of universal dimensions. However, models with more universal dimensions also better predict macaque IT activity and human similarity judgments, suggesting that universality reflects representations relevant to biological vision. These findings have important implications for understanding the emergent representations underlying deep neural network models and their alignment with biological vision. ## Binding Paraphrased Instructions to Encountered Objects in sLLM Agents under Partial Observability DevFeed: [Binding Paraphrased Instructions to Encountered Objects in sLLM Agents under Partial Observability]() Original publisher: [Read original article]() Author: Chanyoung Kim, Harim Song, Byoungchan Ko, Keonyoung Lee, Myungwon Lee, Dahuin Jung Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Large Language Model](), [llm-reasoning](), [rlvr](), [test]() Tags: [agents](), [benchmark](), [cpu](), [cs-lg](), [framework](), [grounding](), [instructions](), [knowledge](), [language-model](), [large-language-model](), [llm](), [llm-agents](), [llms](), [machine-learning](), [object](), [observability](), [persistence](), [proximity](), [robustness](), [tasks](), [training]() ### AI overview ALFWorld-Para finds that controlled instruction paraphrases reduce success rates for RL-finetuned small LLM agents by 21-52%, with object synonyms causing most of the decline. Agents may know what a synonym refers to yet fail to select the matching object; partial observability makes this harder when objects are unseen or previously encountered under another label. The training-free BINDIntent method probabilistically links expressions to observed labels using semantic proximity and estimated persistence relative to unseen candidates. The abstract reports synonym-paraphrase gains of 17.5-49.1 percentage points without extra training or LLM calls, at 0.4 ms CPU overhead per step. For GiGPO-7B, overall success rises from 61.8% to 77.4%, exceeding a prompted 24B LLM. ### Source excerpt arXiv:2611.07599v1 Announce Type: new Abstract: Reliable deployment of LLM agents requires grounding diverse user expressions in their environments. Although RL-finetuned smaller large language model (sLLM) agents rival larger prompted LLMs on embodied tasks, their robustness to instruction paraphrases remains unclear. We introduce ALFWorld-Para to evaluate this robustness under controlled paraphrases, revealing 21-52% declines in success rates (SRs), predominantly driven by object synonyms. Despite retaining knowledge of synonym referents, agents often fail to take the corresponding objects even after observing them, exposing a gap between referential knowledge and object binding. Partial observability further complicates binding, as referents may remain unseen or have already been encountered under different labels. To address this challenge, we propose BINDIntent, a training-free framework that probabilistically binds referring expressions to observed labels using semantic proximity and estimated persistence against unseen candidates. Without additional training or LLM calls, BINDIntent improves synonym-paraphrase success of RL-finetuned sLLMs by 17.5-49.1 percentage points (pp) with only 0.4 ms of CPU overhead per step. It also raises GiGPO-7B's overall SR from 61.8% to 77.4%, surpassing a prompted 24B LLM. Code and benchmark are available at https://github.com/cau-hai-lab/ALFWorld-Para. ## TraceViT: Grounded Trace Supervision for Visual Abstract Reasoning DevFeed: [TraceViT: Grounded Trace Supervision for Visual Abstract Reasoning]() Original publisher: [Read original article]() Author: Binnan Liu, Yechi Ma, Tian Xie, Wei Hua Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [few-shot](), [few-shot-learning](), [generalization in machine learning]() Tags: [abstract](), [abstraction](), [alignment](), [arc](), [code](), [computer-vision](), [cs-ai](), [cs-cv](), [data](), [few-shot](), [grid](), [grounding](), [intermediate](), [iteration](), [model](), [object](), [reasoning](), [supervision](), [trace](), [training](), [transformation]() ### AI overview TraceViT trains a looped visual reasoner to refine ARC grid solutions through intermediate states, addressing the fact that conventional training constrains only the final output. It derives semantically monotonic transformation chains by rewriting and verifying programmatic solutions, then grounds each iteration in a task reference from the demonstrations and a workspace for the current grid. Soft trace alignment preserves the ordering of these states while allowing the number of iterations to vary. TraceViT reports 67.8% pass@2 on ARC-AGI-1 and 24.3% on ARC-AGI-2; ARC-AGI-1 ablations find trace supervision beneficial only when paired with grounding. ### Source excerpt arXiv:2607.29586v2 Announce Type: replace-cross Abstract: The Abstraction and Reasoning Corpus (ARC) tests whether a model can infer an unseen transformation from a few input-output examples and apply it to a new grid. Looped visual reasoners refine predictions over multiple iterations, but conventional training constrains only the final output, leaving intermediate refinements unconstrained. We propose that these refinements should instead follow the transformation step by step. We introduce TraceViT, a looped visual reasoner trained with semantically monotonic transformation chains. We obtain these chains by rewriting and verifying programmatic task implementations, decomposing each solution into intermediate grid states. Each iteration is grounded by a task reference derived from the few-shot demonstrations and an object workspace representing the current grid state. Because these chains may differ in length from the loop, soft trace alignment enforces only their ordering, letting the model allocate iterations freely. TraceViT achieves 67.8% pass@2 on ARC-AGI-1 and 24.3% on ARC-AGI-2. Controlled ablations on ARC-AGI-1 show that trace supervision becomes beneficial only when paired with grounding. Code and data will be available at https://github.com/LiuBinnan/TraceViT. ## Structure from Collision via Hypothesis-Space Reasoning DevFeed: [Structure from Collision via Hypothesis-Space Reasoning]() Original publisher: [Read original article]() Author: Takuhiro Kaneko Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Image](), [Artificial Intelligence]() Tags: [benchmark](), [cs-ai](), [cs-cv](), [cs-gr](), [cs-lg](), [cs-ro](), [event](), [future](), [global](), [gradient](), [hidden](), [hypothesis](), [inspection](), [knowledge](), [local](), [location](), [manipulation](), [matching](), [material](), [motion](), [object](), [optimization](), [reasoning](), [space](), [structure]() ### AI overview The method infers hidden cavities from multiview observations of collision-induced surface motion by treating cavities as explicit structural hypotheses. It first searches the hypothesis space with gradient-free optimization and forward-only material point method (MPM) simulation, then refines the selected hypothesis as a Lagrangian particle system using gradient-based optimization with differentiable MPM. This addresses difficulties with density-based optimization, whose surface-matching gradients do not directly locate cavities and whose density gradients may be weak near voids. On the SfC benchmark, the method improves hidden-structure reconstruction and future prediction over the density-based baseline. The abstract notes that topology stress tests and robustness analyses examine its scope and limitations, but this excerpt gives no specific results from those analyses. ### Source excerpt arXiv:2611.07900v1 Announce Type: cross Abstract: Knowledge of the hidden interior structure of an object, including the presence and location of cavities, is essential for predictive simulation, robotic manipulation, and nondestructive inspection. However, such cavities leave no direct signature on the static visible surface, making collision-induced deformation one of the few cues used to infer their presence. Structure from Collision (SfC) addresses this problem by recovering the hidden interior structure of an object from multiview observations of a collision event. SfC is challenging for two reasons: hidden cavities are only indirectly observable through surface motion, and voids correspond to the absence of material, making direct optimization using standard density representations difficult. The density-based SfC baseline inherits both difficulties. Surface-matching gradients do not directly indicate the cavity location, while density gradients can be weak or uninformative inside or near void regions. Therefore, we reformulate SfC as hypothesis-space reasoning, representing hidden cavities as explicit structural hypotheses whose variables encode available structural assumptions rather than relying on implicit density updates. A global stage performs gradient-free optimization over this hypothesis space using forward-only material point method (MPM) simulation. A local stage then refines the selected hypothesis as a Lagrangian particle system through gradient-based optimization using differentiable MPM simulation. On the SfC benchmark, our method improves hidden-structure reconstruction and future prediction over the density-based SfC baseline. Interior-topology stress tests and robustness analyses characterize its scope and limitations. The project page is available at https://www.kecl.ntt.co.jp/people/kaneko.takuhiro/projects/sfc-hypothesis/. ## Aligning Policy Encoders with a Shape Teacher DevFeed: [Aligning Policy Encoders with a Shape Teacher]() Original publisher: [Read original article]() Author: Madelon Bernardy, Mohammad Mahdi Derakhshani, David M Knigge, Riccardo Valperga, Artem Moskalev Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Robot Manipulation](), [Representation learning](), [Machine learning](), [Image]() Tags: [3d](), [accelerated](), [alignment](), [architecture](), [convergence](), [cs-ai](), [cs-cv](), [cs-ro](), [deployment](), [features](), [image](), [input](), [interface](), [manipulation](), [mechanism](), [object](), [parts](), [patches](), [point](), [policies](), [policy](), [position](), [shape]() ### AI overview Shape-teacher alignment transfers 3D geometric structure into an image-based robotic policy during training, without changing its image-only deployment interface or architecture. It aligns image patches with pretrained Point-MAE features from corresponding objects; the abstract reports more shape- and part-structured image representations, faster policy convergence, and improved success rates on LIBERO. For pretrained policies, convergence was 4.7x faster on the Object suite and 5.6x faster on the Spatial suite. Neither the shape teacher nor depth input is required at deployment. ### Source excerpt arXiv:2611.08275v1 Announce Type: cross Abstract: 3D-aware representations can improve robotic manipulation by providing geometric information that is robust to changes in viewpoint, object position, and occlusion. However, using 3D inputs at deployment requires additional sensing and processing, making image-based policies more practical. We ask whether an image-based policy can acquire 3D shape information during training while retaining a purely image-based interface at deployment. We propose shape-teacher alignment, a training-time mechanism that uses a pretrained 3D shape representation as an alignment target for an image-based policy encoder. Specifically, we use Point-MAE features from individual objects as targets for corresponding image patches, transferring geometric structure to the policy without changing its deployment inputs or architecture. On LIBERO, shape-teacher alignment produces image representations that are more structured around object shape and corresponding parts, while substantially accelerating policy convergence and improving success rates. For pretrained policies, convergence is accelerated by 4.7x and 5.6x on the Object and Spatial suites, respectively, without requiring the shape teacher or depth input at deployment. ## Adaptive Test-Time Memory Allocation for Long-Tailed Human-Object Interaction Detection DevFeed: [Adaptive Test-Time Memory Allocation for Long-Tailed Human-Object Interaction Detection]() Original publisher: [Read original article]() Author: Yuqiu Jiang, Xiaozhen Qiao, Yifan Chen, Ye Zheng, Zhe Sun, Xuelong Li Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [multi-camera tracking]() Tags: [adaptive](), [cache](), [capacity](), [confidence](), [cs-ai](), [cs-cv](), [detection](), [diversity](), [feature](), [framework](), [humanobject-interaction](), [improvements](), [map](), [memory](), [object](), [online](), [prediction](), [predictions](), [results](), [retrieval](), [sample](), [test]() ### AI overview Adaptive Diversity Cache (ADC) improves rare human-object interaction detection by allocating more test-time memory to underrepresented interactions and selecting cache entries using both prediction confidence and local feature diversity. The training-free framework refines predictions from a frozen detector through cache retrieval; it can also fill remaining capacity with selected augmented views. Across three base detectors on HICO-DET and V-COCO, the abstract reports consistent improvements. On HICO-DET, rare-interaction mAP rises by 4.02-8.57 percentage points, with gains also reported for non-rare interactions; online adaptation yields 2.87-7.36 points on rare interactions. These findings are limited to the reported benchmarks and experiments. ### Source excerpt arXiv:2511.18811v3 Announce Type: replace-cross Abstract: Long-tailed human-object interaction (HOI) detection remains challenging because rare interactions are sparsely represented in training data. Although test observations provide additional visual evidence, existing cache-based methods typically assign uniform memory capacity across classes, limiting their ability to capture informative features of underrepresented interactions. To address this issue, we propose Adaptive Diversity Cache (ADC), a training-free, plug-and-play framework that adaptively constructs test-time caches through frequency-aware capacity allocation and confidence-diversity joint sample selection. Specifically, Frequency-Aware Cache Adaptation assigns larger capacities to underrepresented interactions using frequency-dependent priorities and a weighted logarithmic allocation objective, and fills remaining slots with selected augmented views. Confidence-Diversity Joint Cache Selection further combines prediction confidence with local feature diversity to retain informative cache entries. Cache retrieval then refines the predictions of the frozen detector. Experiments on HICO-DET and V-COCO across three base detectors demonstrate consistent improvements. On HICO-DET, ADC improves mean average precision (mAP) for rare interactions by 4.02--8.57 percentage points, while also improving performance on non-rare interactions. Online adaptation further achieves gains of 2.87--7.36 percentage points on rare interactions. These results demonstrate that adaptive test-time memory effectively complements existing HOI detectors and improves long-tailed detection performance. ## Lakebase Postgres is built on object storage, but 99.99999% of reads never touch it DevFeed: [Lakebase Postgres is built on object storage, but 99.99999% of reads never touch it]() Original publisher: [Read original article]() Author: Carlota Soto Published: 2026-11-05T12:00:00Z Content type: article Language: en Sources: [Neon Docs]() Topics: [Amazon S3](), [Cache](), [data lake]() Tags: [2](), [3](), [about](), [accessible](), [agents](), [all](), [also](), [analytics](), [architecture](), [argument](), [but](), [conclusion](), [engineering](), [isn-t](), [lakebase](), [object](), [object-storage](), [postgres](), [right](), [running](), [storage]() ### AI overview Lakebase Postgres keeps object storage off the active query path through layered caching: the compute node serves about 99.46% of page accesses from its local cache, pageserver NVMe serves about 0.54%, and object storage handles roughly 39 requests per second--about one in 28 million page accesses. Each tier caches data fetched from the tier below. Pageservers answer GetPage@LSN requests by combining WAL deltas with image layers; after base pages were created more eagerly, p99 replay fell to around six deltas, while p90 stayed below one. Object storage fetches mainly occur on first reads after long inactivity or when accessing old branch or restore points; subsequent reads use caches. The architecture also separates compute from durable storage: safekeepers replicate WAL, pageservers persist history asynchronously, and object storage holds immutable history, enabling scale-to-zero, branching, and restores without copying data. ### Source excerpt The most common argument against running Postgres on object storage is that a Postgres backed by object storage must be a slow Postgres. The premise is right, but the conclusion isn't. ## Touch\\'{e}World tech report : Towards Online Unified Camera--Hand Exploration in Generative Worlds DevFeed: [Touch\\'{e}World tech report : Towards Online Unified Camera--Hand Exploration in Generative Worlds]() Original publisher: [Read original article]() Author: Jiaao Yu, Yanjian Zhang, Haoyang Jiang, Guillaume Wisniewski, Nadi Tomeh, Shijie Jia Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [World models](), [Generative AI](), [peft]() Tags: [camera](), [continuous](), [cs-ai](), [cs-cv](), [exploration](), [generation](), [generative](), [image](), [interaction](), [model](), [models](), [motion](), [navigation](), [object](), [online](), [spatial](), [temporal](), [training-data](), [video](), [world-model](), [world-models](), [worlds]() ### AI overview TouchéWorld combines camera navigation and continuous hand gestures to control video generation in a virtual environment. Given an initial image and scene prompt, it conditions generated video on temporally aligned camera and hand controls. The framework builds joint training data from egocentric videos using annotation completion, consistency filtering, and temporal alignment, then adapts a pretrained camera-controlled world model with hand-skeleton and numerical motion features. Causal adaptation and few-step distillation enable chunk-level online generation, so users can adjust controls in response to visual feedback. Its TouchéBench evaluates hand-motion following, camera-hand coordination, online interaction, and object response. ### Source excerpt arXiv:2611.06061v1 Announce Type: cross Abstract: Recent video world models allow users to explore generated environments through camera navigation. Continuous hand motion provides a fine-grained action channel for interacting with the people and objects encountered during exploration. We present Touch\'{e}World, a framework that integrates gesture-driven interaction into a camera-controlled video world model. Given an initial image and a scene-description prompt, the model generates video conditioned on temporally aligned camera and hand controls. We construct joint camera--hand training data from heterogeneous egocentric videos through annotation completion, consistency-based filtering, and temporal alignment. Building on a pretrained camera-controlled world model, we combine spatial hand-skeleton conditioning with numerical hand-motion features through parameter-efficient adaptation. Causal adaptation and few-step distillation support chunk-level online generation, allowing users to update their controls based on generated visual feedback. We further introduce Touch\'{e}Bench to evaluate hand-motion following, camera--hand coordination, online interaction, and object response. Together, these components establish a framework for studying continuous exploration and hand-driven interaction within generative worlds. ## MonoLaDN: Language-Guided DeNoising for Monocular 3D Object Detection DevFeed: [MonoLaDN: Language-Guided DeNoising for Monocular 3D Object Detection]() Original publisher: [Read original article]() Author: Kiet Dang Vu, Cao Le Minh Khoa, Nguyen Viet Tri Pham, An Nguyen Gia, Tam Le Thuc Ho, Duc Dung Nguyen Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [vlm]() Tags: [3d](), [cs-ai](), [cs-cv](), [denoising](), [detection](), [dimension](), [domains](), [experiments](), [features](), [framework](), [image](), [inference](), [knowledge](), [language](), [language-models](), [latency](), [model](), [models](), [noise](), [object](), [object-detection](), [semantic]() ### AI overview MonoLaDN transfers a vision-language model's geometric priors into a monocular 3D detector during training, while keeping inference vision-only. A frozen VLM uses images and ground-truth boxes to produce structured reasoning about object occlusion, depth, dimensions, and angle. The detector is trained to denoise perturbed reasoning queries and reconstruct the clean reasoning, encouraging its depth-guided features to retain those priors. The reasoning path is removed at inference, so the deployed model adds no latency or parameters. The abstract reports consistent gains over state-of-the-art detectors on KITTI 3D and nuScenes, and generalization to new domains; it provides no specific metrics or domain details. ### Source excerpt arXiv:2611.06282v1 Announce Type: cross Abstract: Monocular 3D object detection offers a low-cost route to 3D perception, but it remains fundamentally ill-posed, since depth must be inferred from a single view. Recent transformer-based detectors learn geometry purely from image features and 3D box annotations, lacking the high-level semantic priors that vision-language models (VLMs) acquire from large-scale pretraining. VLMs offer such priors, but deploying one directly as a detector is impractical due to its heavy inference cost. We propose MonoLaDN, a language-guided denoising framework that transfers the geometric reasoning of a VLM into a monocular 3D detector while keeping the deployed model purely vision-based. During training, a frozen VLM produces structured reasoning about each object's occlusion, depth, dimension, and angle from the image and its ground-truth boxes. We encode this reasoning into queries, perturb them with noise, and train the detector to reconstruct the clean reasoning. Reconstructing the corrupted reasoning forces the detector's depth-guided features to internalize the VLM's geometric priors, so this knowledge remains available even after the reasoning path is discarded at inference, adding no latency or parameters to the deployed model. Experiments on KITTI 3D and nuScenes show that MonoLaDN consistently outperforms state-of-the-art detectors and generalizes well to new domains. ## Clearing up a robot's fuzzy memory DevFeed: [Clearing up a robot's fuzzy memory]() Original publisher: [Read original article]() Author: Adam Zewe | MIT News Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [MIT AI News]() Topics: [Robot Manipulation](), [Image]() Tags: [aeronautical-and-astronautical-engineering](), [algorithms](), [artificial-intelligence](), [computer-science-and-technology](), [computer-vision](), [electrical-engineering-and-computer-science-eecs](), [environment](), [harry-zhang](), [laboratory-for-information-and-decision-systems-lids](), [luca-carlone](), [machine-learning](), [memory](), [mit-schwarzman-college-of-computing](), [nicolas-gorlo](), [object](), [research](), [robot](), [robotics](), [school-of-engineering](), [spatio-temporal-memory](), [system](), [u-s-armed-forces](), [uncertainty-quantification](), [vision-language-models]() ### AI overview UQ-DAAAM helps a robot detect and correct unreliable object descriptions in its map-based memory. It combines the vision-language model's confidence in a caption with its consistency against descriptions from other viewpoints. When uncertainty crosses a threshold, a heuristic selects useful additional views without rerunning the vision-language model; a language model then combines captions, favoring newer and less ambiguous ones. The researchers report that this resolves ambiguities for about 92% of objects using two additional captions per object, raises description accuracy by 5% over the state of the art, and improves answers about objects, locations, and when they were seen compared with baseline methods. The excerpt describes planned extensions to other robots and uncertainty-guided revisits; it does not report those as completed. ### Source excerpt A new system identifies and corrects unreliable object descriptions stored in a robot's memory, improving the robot's ability to interact with its environment. ## HTCE: Homeomorphism-inspired Topological Cross-Scene Experience Canonicalization and Reuse for Object Navigation DevFeed: [HTCE: Homeomorphism-inspired Topological Cross-Scene Experience Canonicalization and Reuse for Object Navigation]() Original publisher: [Read original article]() Author: Xiaoran Xu, Yupeng Wu, Tianyu Xue, Xuanran Dong, Yifan Xu, Xiaoshan Yang, Changsheng Xu Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Robot Navigation](), [Artificial Intelligence]() Tags: [canonicalization](), [cross](), [cs-ai](), [experience](), [feature](), [framework](), [graph](), [guidance](), [language](), [matching](), [navigation](), [object](), [performance](), [policy](), [region](), [retrieval](), [reuse](), [route](), [semantic](), [sequences](), [stored]() ### AI overview HTCE reuses successful object navigation experience across residential scenes by reducing differing feature and topology subdivisions into comparable semantic sequences and structural signatures. Its constraints preserve meaningful anchors, branches, and endpoints; subdivision-aware matching then retrieves prior approaches to guide a frozen vision-language policy. The authors report lossless index compression merging 17.7-23.6% of route descriptions without reducing task performance. With the same frozen vision-language backbone across compared systems, HTCE achieved the best reported success rate (SR) and success weighted by path length (SPL) in each setting: 59.9%/31.8% on HM3D v0.1, 71.0%/31.5% on HM3D v0.2, and 46.6%/16.6% on MP3D. Reported SR gains reached 4.9 points, while structured retrieval contributed up to 2.8 SPL points. ### Source excerpt arXiv:2611.04836v1 Announce Type: new Abstract: As embodied agents perform object-goal navigation across residential environments, they accumulate successful experiences whose semantic structure recurs across scenes even when geometric trajectories do not. Reusing such experience is challenging because differences in region naming and observation granularity produce distinct graph representations of similar approaches, while meaningful differences in anchors, branches, and endpoints must stay distinguishable. We introduce HTCE, a homeomorphism-inspired topological framework that canonicalizes and reuses navigation experience. Feature and topology subdivisions characterize representational variations, and their semantically constrained reductions produce comparable semantic sequences and structural signatures while preserving semantic anchors, branching relations, and navigation endpoints. HTCE then retrieves historical successful approaches through subdivision-aware matching as guidance for a frozen vision-language policy. Canonicalization compresses the experience index losslessly, merging 17.7--23.6\% of distinct stored route descriptions while retaining full task performance. Under a single frozen vision-language backbone shared by all compared systems, HTCE attains the best SR and SPL in every setting---59.9\%/31.8\% on HM3D v0.1, 71.0\%/31.5\% on HM3D v0.2, and 46.6\%/16.6\% on MP3D---with SR gains of up to 4.9 points and structured retrieval contributing up to 2.8 SPL points. The code is available in https://github.com/startracker0/htce.git ## Do Agentic Policies Need to Look at Every Step? Geometric Abstraction for Reasoning and Action DevFeed: [Do Agentic Policies Need to Look at Every Step? Geometric Abstraction for Reasoning and Action]() Original publisher: [Read original article]() Author: Yihang Liu, Yiming Chen, Mingzhe Zheng Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [llm-reasoning](), [Multi-step AI processes](), [abstraction]() Tags: [3d](), [abstraction](), [agentic](), [completion](), [continuous](), [cs-ai](), [cs-cv](), [cs-ro](), [execution](), [images](), [interaction](), [interface](), [language](), [language-models](), [manipulation](), [metric](), [models](), [multimodal](), [object](), [offline](), [online](), [parts](), [reasoning]() ### AI overview The paper proposes shifting routine robot spatial reasoning from repeated image interpretation to grounded text and symbols. Its 3D interface uses typed, evidence-aware references to object parts, grounds them in calibrated RGB-D, and provides compact textual checkpoints with metric residuals after actions. Symbolic task structure can be reused across decisions while metric values are rebound to the current scene, with direct visual reasoning used on demand. The abstract reports improved multi-step manipulation completion and fewer repeated multimodal queries in offline policy synthesis and continuous online execution. ### Source excerpt arXiv:2611.05058v1 Announce Type: cross Abstract: Recent multimodal language models can reason directly about robot scenes, but it is unclear whether they need to repeatedly reinterpret raw visual observations at every step of an interaction. We propose shifting routine spatial reasoning from images to text and symbols through a 3D geometric abstraction that is refreshed locally from observations and read directly by the language models. Specifically, our interface represents scene geometry via typed, evidence-aware symbolic references to object parts, grounds them from calibrated RGB-D, and returns compact textual checkpoints with metric residuals after each action. Symbolic task structure can thus be reused across decisions while metric values are rebound to the current scene, with direct visual reasoning invoked only on demand. Across offline policy synthesis and continuous online execution, reasoning over grounded text and symbols improves multi-step manipulation completion while substantially reducing repeated multimodal queries, positioning explicit geometric state as an effective substrate for agentic robot policies. ## More Accurate, Less Human: Gestalt Grouping in Vision Models DevFeed: [More Accurate, Less Human: Gestalt Grouping in Vision Models]() Original publisher: [Read original article]() Author: Sudhanva Manjunath Athreya, Sai Phani Kumar Malladi Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Artificial Intelligence](), [image recognition](), [color]() Tags: [alignment](), [auditing](), [battery](), [benchmark](), [color](), [cs-cv](), [cs-lg](), [design](), [human-data](), [metrics](), [models](), [object](), [organization](), [recognition](), [series](), [tasks](), [training](), [vision](), [visualization]() ### AI overview The study finds that benchmark accuracy can overstate how closely vision models organize visual content like people do. It tests 45 models across five training families on four Gestalt grouping tasks, scoring their responses against results from prior human perception studies. Human-response alignment distinguishes perceptual organization that conventional performance metrics miss; several closed models show substantially lower alignment than their benchmark accuracy suggests. The authors present this comparison as a reusable way to audit models used in visualization, without conducting a new user study. ### Source excerpt arXiv:2608.10195v2 Announce Type: replace-cross Abstract: Human vision organizes what it sees into wholes: same-colored points group into series, similar marks cohere into categories, and shapes complete into recognizable objects. These are the Gestalt operations that visualization design builds on. Whether vision models organize visual content this way has not been systematically tested. We introduce a behavioral battery that scores models against human data from prior perception studies on four grouping tasks: mark-color odd-one-out, color-series counting, silhouette recognition, and object odd-one-out. We apply it to 45 models across five training families: supervised, self-supervised, and contrastive vision-language encoders, open-weight VLMs, and closed foundation models. The battery reveals that agreement with human responses captures aspects of perceptual organization that conventional performance metrics fail to distinguish, with several closed models exhibiting substantially lower alignment than their benchmark accuracy would suggest. Scoring against published perception data therefore gives visualization research a reusable yardstick, requiring no new user study, for auditing whether the models now entering visualization pipelines organize what they see the way their human audience does. ## Where Did the Saved Bytes Go? Compact Object Headers in IntelliJ IDEA DevFeed: [Where Did the Saved Bytes Go? Compact Object Headers in IntelliJ IDEA]() Original publisher: [Read original article]() Author: Igor Kulakov Published: 2026-11-03T12:20:40Z Content type: article Language: en Sources: [JetBrains]() Topics: [IntelliJ IDEA](), [memory handling]() Tags: [boundary](), [garbage-collection](), [headers](), [idea](), [intellij-idea](), [intellij-platform](), [java](), [jvm](), [memory](), [object](), [performance]() ### AI overview In tests inside IntelliJ IDEA, compact object headers reduced modeled heap footprints by about 6.3% for the Welcome screen and Petclinic objects and 8.2%-8.4% for monorepo objects, but produced no consistent RSS reduction at identical -Xmx. They also cut allocated volume by 8.8%-13.0%; monorepo indexing runs recorded fewer GC events and lower mean collection time and CPU, while scripted operation timings suggested only a modest speedup. The minimum tested heap that completed every repeat fell 6% for Petclinic and 17% for the monorepo; these are workload thresholds, not production recommendations. Savings depend on object shapes, fields and alignment, so smaller headers do not shrink every object or guarantee lower RSS or faster execution. Measure heap, allocation, GC, performance and heap limits separately. ### Source excerpt AI agents may make code faster to produce, but they don't make it cheaper to run. In applications managing millions of objects, memory allocation and garbage collection still matter. The JVM has been addressing these costs on several fronts: ZGC targets garbage collection pauses, Project Leyden tackles startup and warmup, and Project Lilliput reduces the [...] ## A Tool in the Forward Pass: Frozen Tape Modules as Layers of a Small Language Model DevFeed: [A Tool in the Forward Pass: Frozen Tape Modules as Layers of a Small Language Model]() Original publisher: [Read original article]() Author: Michael Granitzer, Saber Zerhoudi, Mohammad Al-Maamari, Hai Dang Dang, Kanishka Ghosh Dastidar Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Large Language Model](), [LLM evaluation / benchmarking](), [Neural Network](), [model architecture]() Tags: [arithmetic](), [bounded](), [cs-ai](), [cs-ne](), [language-model](), [layers](), [module](), [object](), [symbols](), [tape](), [tracking](), [use-case]() ### AI overview The paper tests frozen neural modules that run inside a language model's forward pass: a learned writer encodes computation as a short symbolic tape, the frozen module reduces it to a result, and a learned reader returns that result. In gemma-3-1b-it, a writer trained on labelled tapes gets all 750 BIG-Bench Hard shuffled-object tracking items correct; at seven people it uses about one tenth the GPU kernel time of a text-tool baseline with the same host model and code, at accuracy 1.00 versus 0.99. Learning the writer from answers alone with a hand-written copying prior is less reliable: it succeeds in 5 of 6 runs with the 1B host and 2 of 4 with the 4B host. Two independently trained layers retain task accuracy when combined, and a learned gate stays silent on tested general-text and look-alike prompts. The evidence suggests a narrow benefit for bounded computations returning one symbol; reliable input extraction and deciding when to act remain open problems. ### Source excerpt arXiv:2611.00783v1 Announce Type: new Abstract: Calling a tool through generated text adds decoding steps, even when the computation itself is short. We study an alternative: a small neural module that runs inside a language model's forward pass. A learned writer turns hidden states into a symbolic tape, a short sequence of discrete symbols that encodes the computation. The module, trained separately and then frozen, reduces the tape to a result, and a learned reader returns that result to the model. We evaluate arithmetic and object tracking in gemma-3-1b-it. With a writer trained on labelled tapes, the layer answers all 750 items of BIG-Bench Hard's tracking shuffled objects correctly. At seven people it needs about 1/10 of the GPU kernel time of a text-tool baseline that uses the same host model and exact code, with accuracy 1.00 against 0.99. The writer can also be learned from answers alone, guided by a hand-written copying prior, but less reliably: 5 of 6 runs succeed in the 1B host and 2 of 4 in the 4B host. Two independently trained layers keep their task accuracy when attached to one model, and a learned gate keeps them silent on the tested general-text and look-alike prompts. The results support a narrow use case: a bounded computation that returns one symbol can avoid the decoding cost of a textual tool call. Extracting the right input reliably, and deciding when to act, remain the main open problems. ## Latent Spectral Prompting for Multimodal Hyperspectral Object Tracking DevFeed: [Latent Spectral Prompting for Multimodal Hyperspectral Object Tracking]() Original publisher: [Read original article]() Author: Wangshu Cai, Jun Zhou, Lei Wang, Mohammad Aminul Islam, Xu Han, Kuldip Paliwal Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Representation learning](), [Image processing]() Tags: [abstract](), [all](), [annotations](), [arxiv](), [autoencoder](), [band](), [based](), [best](), [bridges](), [cs-ai](), [cs-cv](), [data](), [domain](), [domains](), [encoding](), [experiments](), [foundation](), [foundation-models](), [multimodal](), [object](), [prompting](), [tracking]() ### AI overview SProTrack transfers temporal modeling from pretrained RGB trackers to hyperspectral object tracking while learning spectral adaptation from unlabeled data through self-supervision. An autoencoder represents each full hyperspectral frame as a global, scene-level latent representation that guides spectral prompt extraction; the RGB tracker retains temporal modeling on RGB-compatible inputs. The authors contrast this with band selection, which discards spectral information, and prompt methods based on local-search crops. They report best overall performance on HOT2020 and HOT2023, with strong results across spectral domains. The supplied abstract does not provide quantitative results or implementation details. ### Source excerpt arXiv:2611.02512v1 Announce Type: cross Abstract: Hyperspectral videos provide rich spectral signatures for robust tracking under challenging conditions such as camouflage and low illumination. However, limited hyperspectral annotations and the large RGB-hyperspectral domain gap hinder the direct use of modern temporally pretrained RGB trackers. We propose SProTrack, a multimodal framework that transfers temporal knowledge from pretrained RGB trackers to the hyperspectral domain through latent spectral prompting. The key idea is to decouple temporal modeling from spectral adaptation, allowing temporal dynamics to be inherited from RGB foundation models while hyperspectral information is learned from unlabeled data in a self-supervised manner. Unlike band-selection methods that discard most spectral information or prompt-based approaches that rely on local-search crops, SProTrack learns a global, scene-level spectral representation from the full hyperspectral frame using an autoencoder. This latent representation is used to guide spectral prompt extraction while a pretrained RGB tracker preserves strong temporal modeling on RGB-compatible inputs. By encoding spectral-spatial structure into compact latent prompts, SProTrack bridges the RGB-hyperspectral modality gap without extensive retraining or dense supervision. Experiments on HOT2020 and HOT2023 demonstrate that SProTrack achieves the best overall performance, with strong results across all spectral domains. ## GeoGuesser Bench: Benchmarking Vision-Language Agents on Interactive Geolocation DevFeed: [GeoGuesser Bench: Benchmarking Vision-Language Agents on Interactive Geolocation]() Original publisher: [Read original article]() Author: Huanzhi Mao, Yichuan Wang, Michael Koo, Wenyuan Huang, Huy Doan, Jinghan Xia, Ziyao Wang, Joseph E. Gonzalez Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [vlm](), [Artificial Intelligence](), [rlvr](), [Agent Harness]() Tags: [agents](), [bench](), [benchmarking](), [cs-ai](), [cs-lg](), [environment](), [exploration](), [geolocation](), [interaction](), [interactive](), [knowledge](), [location](), [model](), [navigation](), [object](), [offline](), [training](), [vision](), [vision-language-model](), [vlm](), [vlms](), [zoom]() ### AI overview GeoGuesser Bench evaluates how vision-language agents combine world knowledge with exploration across 282 tasks in an offline Street View environment, where they can rotate, zoom, and move between panoramas. Tasks cover coordinate guesses, place naming, and finding, reading, or counting nearby objects; 42 coordinate tasks come from GeoGuessr World Championship rounds for comparison with expert guesses. Across 23 models, the best reported accuracy is 62.5% for place naming and 35.0% for object finding. The top agent averages 4352.4 championship points, below the better expert's 4770.3 average per round. Localization and exploration performance correlate only weakly, and failures are categorized as vision, navigation, or tool-use errors. Strong agents such as Claude Opus 4.8 remain capable with limited views, while weaker agents degrade. A Planner-Worker-Compressor harness reduces planner tokens by 57-61% by compacting interaction history. ### Source excerpt arXiv:2611.02698v1 Announce Type: new Abstract: Vision-language model (VLM) agents that act in the open world need two abilities: world knowledge to interpret what they see, and exploration to find what they cannot yet see. We study how agents combine the two through interactive visual geolocation. GeoGuesser Bench contains 282 tasks in an offline Street View environment where agents rotate, zoom, and move between real panoramas. Its three task types shift the balance from knowledge to exploration: agents (1) report the coordinates of their starting location, (2) name the enclosing street, city, state, or country, or (3) find, read, and count nearby objects. The coordinate tasks in (1) also include 42 real GeoGuessr World Championship rounds, whose expert guesses provide a human reference. We evaluate 23 VLMs: the best accuracy is 62.5% on place naming and 35.0% on object finding, and the best agent averages 4352.4 GeoGuessr points on the championship rounds, against 4770.3 for the better expert in each round. Beyond these headline numbers, world knowledge and exploration are only weakly correlated across models: agents that localize well do not necessarily explore well. To understand where agents fail, we categorize their errors into vision, navigation, and tool-use failures. Further ablations show that the strongest agents, such as Claude Opus 4.8, still perform well with limited views, while weaker agents degrade; customizing the environment thus adjusts task difficulty. The environment's verifiable rewards and restorable state also support training, and our Planner-Worker-Compressor harness cuts planner tokens by 57-61% by compacting the interaction history. We hope GeoGuesser Bench and its environment serve as a high-quality foundation for simulating real-world exploration at scale. ## Detection Instead of Inversion: T1-T2 Microstructure Quantification With Detection Transformers DevFeed: [Detection Instead of Inversion: T1-T2 Microstructure Quantification With Detection Transformers]() Original publisher: [Read original article]() Author: Sebastian Endt, Fatih \\"Ozkan, Marion I. Menzel Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Transformers](), [Neural Network]() Tags: [abstract](), [approach](), [arxiv](), [challenge](), [code](), [correlation](), [cs-cv](), [cs-lg](), [data](), [design](), [detection](), [faster](), [full](), [instead](), [magnitude](), [model](), [object](), [object-detection](), [orders](), [physics-med-ph](), [precision](), [transformers](), [with]() ### AI overview The study replaces voxel-wise inverse Laplace reconstruction of T1-T2 spectra with a Detection Transformer that maps 64 TI/TE signals directly to compartments described by T1, T2, and signal fraction. On simulated data, it reports mAP@10 of 76.8%, compared with 53.8% for the authors' previous model and 66.1% for regularized NNLS. DETR reportedly outperforms NNLS in almost every tested setting across compartment count, signal fraction, and SNR, with reconstruction orders of magnitude faster. The supplied abstract does not establish performance on measured data. ### Source excerpt arXiv:2611.01790v1 Announce Type: cross Abstract: T1-T2 correlation MRI has strong potential to quantify microstructural compartments without imposing a biological model. Traditionally, full T1-T2 spectra per voxel are reconstructed with an ill-conditioned and time-consuming Inverse Laplace Transform. We propose to view this challenge as an object detection task: A Detection Transformer (DETR) takes 64 TI/TE signals and directly predicts a set of compartments specified by T1, T2, and signal fraction. We achieve a mean average precision mAP@10 of 76.8% on simulated data, outperforming both our previous model design (53.8%) and regularized NNLS (66.1%). We show that our DETR approach outperforms NNLS in almost every setting across compartment count, compartmental signal fraction, and SNR. Reconstruction is orders of magnitude faster than NNLS. The results are fully reproducible from our public code: github.com/SebastianEndtTHI/t1t2-detr ## SCOUT: Planning under Occlusion via Object-Centric World Model Rollouts DevFeed: [SCOUT: Planning under Occlusion via Object-Centric World Model Rollouts]() Original publisher: [Read original article]() Author: Amir Hossain Raj, Xuesu Xiao Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Robot Manipulation](), [monte carlo tree search](), [World models](), [Large Language Model]() Tags: [3d](), [cs-ai](), [cs-cv](), [cs-ro](), [evaluations](), [llm](), [manipulation](), [monte-carlo-tree-search](), [object](), [planning](), [real-world](), [reasoning](), [vision-language-action](), [world-model]() ### AI overview SCOUT plans manipulation in clutter by reconstructing hidden geometry from an object-centric scene representation and encoding blocker hierarchies and accessibility corridors in a relational state. An LLM proposes structured actions, which Monte Carlo Tree Search evaluates using a learned action-conditioned world model. The abstract reports that SCOUT outperforms VLA policies, embodied agents, and 3D world models on Reveal and Placement tasks, with larger gains on long-horizon tasks involving occluded geometry, and describes a real-world demonstration in a cluttered shelf environment. ### Source excerpt arXiv:2611.01600v1 Announce Type: cross Abstract: Robots manipulating in cluttered spaces must often plan from partial views where foreground objects occlude the target, surrounding geometry, and usable free space. Reactive policies often miss these hidden constraints, while exhaustive search becomes computationally prohibitive in dense clutter. We present SCOUT, an occlusion-aware planning framework that bridges high-level deliberation with grounded physical rollouts. SCOUT leverages an object-centric scene representation to reconstruct hidden geometry and derive a relational planning state capturing blocker hierarchies and accessibility corridors. To navigate the vast action space in clutter, an LLM proposes structured action candidates which are then rigorously evaluated via Monte Carlo Tree Search (MCTS) using a learned action-conditioned world model. Evaluations on 'Reveal' and 'Placement' tasks show that SCOUT outperforms Vision-Language-Action (VLA) policies, embodied agents, and 3D world models, with significant gains in long-horizon tasks requiring reasoning over occluded geometry. We further validate the framework's robustness through real-world robotic manipulation in a cluttered shelf environment. [Next page]()