#01Oct 30, 2026
cs.RO
REA: A Routed Embodied Agent Runtime for Asynchronous Brain-Cerebellum Control
Pengfei Zhou, Yixing Ma, Zhengfeiyang Zhang and 5 more
Foundation models are becoming remarkably reliable at high-level reasoning, yet they remain too slow and costly to serve as fine-grained controllers. Conversely, fast policies like VLAs and whole-body controllers offer more efficient reflexes, but struggle to generalize. While pairing a deliberative brain (foundation models) with a reactive cerebellum (fast policies) seems natural for embodied systems, two obstacles remain: methods are built and scored on incompatible stacks, and their asynchronous coordination is difficult. For example, an agent that calls the VLA in the agentic loop can stall the fast execution, while a VLA that calls the agent on errors tends to re-trigger fallbacks repeatedly or completely miss silent failures. To address this, we propose REA (Routed Embodied Agent), a unified runtime where a deliberative System 2, an embodied router System 1 and reactive System 0 policies cooperate under a single golden rule - no running policy waits on the brain. Instead of querying the brain at each failure, REA employs a route memory that stores each policy's measured success ahead of time, a boundary-triggered judge gate that consults the brain on policy execution, and a shared-budget escalation ladder for bounded recovery. Evaluated on 3,000 test episodes across 60 LIBERO-Pro tasks, REA reaches 95.9% success and beats the best single-agent baseline in its own policy pool (94.7%) while reducing control steps by 25% and control time by 17%. In a separate 100-episode RoboTwin 2.0 study, REA achieves 84% success with VLA policies and 86% with VLA+code, compared with 80% for the strongest single VLA baseline MEDex. These results demonstrate learned and programmatic policies operating within the same runtime. REA also transfers to BEHAVIOR-1K and a real Unitree G1 humanoid, achieving the highest observed success among the compared methods.
#02Oct 30, 2026
cs.CV
Nereus: A Large-Scale Underwater Dataset for Fine-Grained Attribute Understanding and Grounded Counting Perception
Jingtao Deng, Zihao Jin, Yuehe Chen and 3 more
Underwater imagery records rich ecological information and supports applications such as science communication, fishery monitoring, ecological conservation, and oceanographic research. Beyond general scene understanding, these applications require models to recognize fine-grained biological traits and reason over spatially grounded evidence. However, existing underwater datasets often provide coarse category labels, generic captions, or total-count supervision, which limits their use for organism-level analysis and ecological reasoning. We introduce Nereus, a large-scale dataset for Underwater Fine-grained Perception that jointly supports Fine-Grained Object Attribute Understanding and Grounded Counting Perception. Nereus contains 88K images and 2.68M question-answer pairs, including annotations for 136K marine organism objects across 2,690 species and 1,515 genera, as well as 65K point annotations and 67K grounded counting QA pairs. Experiments with representative multimodal models show that Nereus improves downstream adaptation over existing underwater-domain data and provides a foundation for future research on fine-grained marine perception.
#03Oct 30, 2026
cs.CV
Rethinking Asymmetry in Self-Supervised 3D Geometry Learning
Gyeongjin Kang, Freeman Cheng, Botao Ye and 3 more
Self-supervised learning relies on asymmetry between what a model observes and what it must explain. Recent methods learn 3D geometry from unposed images by predicting cameras and scene structure, then rendering held-out views. To avoid trivial solutions, they impose this asymmetry in the input space, restricting how views interact through separate branches or masked attention. We revisit this design with geometric accuracy, rather than rendering quality, as the main objective. When a held-out view lies between context views, the context already constrains the geometry it observes. We therefore impose the asymmetry in the 3D representation space, excluding held-out views only from the reconstructed scene while a shared backbone uses all views for correspondence. This improves depth and pose estimation over input-space designs. We further exploit view density: sparse and dense observations depict the same scene but provide different amounts of correspondence. Aligning sparse-context features with dense-context features supplies a geometric training signal alongside rendering, without a pretrained geometric teacher. We additionally align features with a frozen vision foundation model, which brings further gains. None of these signals requires geometric annotation, and the resulting model improves geometry estimation over prior self-supervised methods. Controlled experiments isolate the contribution of each component, and self-supervised pretraining also improves subsequent supervised training.
#04Oct 30, 2026
cs.CV
A Patch is More Than a Token
Zixiong Wang, Haiping Ma
Vision Transformers represent each image patch through a canonical token produced under one observation condition. That token is a powerful interface, but it is still one realized state of the patch. We introduce PMTT, which establishes two foundations: a patch representation object can extend from one canonical state to a canonical-centered Representation State Family, and identity-preserving variation opens Patch-Level Local Feature Space. We develop four parts of this representation program. Patch Information Acquisition and Dense Orbital Views expose lawful same-patch states. Visible-to-Hidden prediction tests whether observed states reveal structure beyond themselves. Intrinsic analysis identifies low-dimensional, energy-concentrated, patch-specific, predictive geometry. Representation Organization separates acquiring more information from organizing that information, with an analytic Manifold as the central structured representation. We realize PMTT on frozen DINOv3 S/B/L/H+ backbones without neural post-training of the representation. Across two independent 4,096-anchor cohorts, Manifold explains 70.8-79.6% of Hidden variation, compared with 55.0-72.3% for directly Observed states, and wins on more than 95% of anchors at every scale. Effective rank falls from about 2.5 to 2.0 while dominant energy concentration rises with backbone scale. Across six visual capabilities and four scales, Observed exceeds the canonical Center in 21 of 24 conditions, Manifold exceeds Center in 23, and Manifold exceeds Observed in 16. The strongest current consequence appears in physical correspondence, where MegaDepth satisfies Center < Observed < Manifold at every scale. These results show that pretrained visual backbones contain richer patch information than the standard single-token interface directly exposes. Code and experimental assets are available at https://github.com/zixiongwang-ai/PMTT.
#05Oct 30, 2026
cs.CV
TwinPast: Benchmarking Historical Evidence Use in Streaming Video Understanding with Evidence Contrastive Video Pairs
Haotian Yang, Yong Chen, Cheng Peng and 4 more
High accuracy on streaming video benchmarks does not establish whether a model uses visual history. Our review of OVO-Bench and StreamingBench finds that recent frames suffice for 32% of sampled historical questions, while models also answer many questions correctly from the question and options alone. We introduce TwinPast to test whether answers correctly follow changes in history, using 350 human-reviewed video pairs spanning five historical tasks. Each pair holds the recent video context, question, and ordered options fixed while changing critical historical evidence and the correct answer. Each video is evaluated independently, and PairAcc requires correct answers for both histories. We evaluate 34 model and input configurations. For Qwen3-VL-8B and DeepSeek-Flash, historical inputs improve performance on TwinPast, reversing the recent-frame advantage on existing benchmarks. The strongest evaluated streaming system reaches 49.14% PairAcc, compared with 66.57-91.43% for the evaluated general-purpose models. This gap persists across all five tasks. Case studies further show that models select states from the wrong time or contradict recorded event order despite retaining the relevant evidence. These findings highlight the need to connect historical memory and retrieval with reasoning that uses evidence according to the question's temporal references and event relations.