ReactVLA: Fast and Lightweight Reactive Robot Manipulation via Improved Mean Flow Action Generation

Diffusion-based Vision-Language-Action (VLA) policies have demonstrated strong capability in modeling expressive and multimodal action distributions. However, their reliance on iterative sampling introduces substantial inference latency, which limits their applicability to reactive closed-loop robot manipulation. To address this limitation, we propose \texttt{ReactVLA}, a lightweight and low-latency VLA framework for real-time robotic manipulation. \texttt{ReactVLA} combines two complementary designs: (1) an improved Mean Flow (iMF) action generator that reduces expensive multi-step diffusion sampling to one-to-few-step action generation, and (2) Attention Residuals (AttnRes), a dynamic depth-wise feature routing mechanism that replaces uniform residual accumulation to better preserve task-relevant multimodal representations. We evaluate \texttt{ReactVLA} on large-scale simulation benchmarks, including LIBERO and RoboIMI, as well as real-world robotic manipulation tasks. Experimental results show that \texttt{ReactVLA} consistently outperforms similarly sized VLA baselines, including SmolVLA and $π_0$. On challenging precision manipulation tasks, \texttt{ReactVLA} achieves up to a 1.65$\times$ improvement in task performance while providing more than a 4$\times$ increase in inference speed compared with leading VLA models. Finally, it reduces real-world policy latency to below 38.6 ms, enabling fast reactive control on physical robot platforms. Please check out our project website at: https://game-loader.github.io/ReactVLA/.

LoopMoE: Unifying Iterative Computation with Mixture-of-Experts for Language Modeling

Mixture-of-Experts (MoE) and looped architectures scale models along two orthogonal axes, namely parameter capacity and effective depth. However, mainstream looped architectures rely on dense backbones that couple parameter count with per-token FLOPs, which makes it impossible to isolate the effect of iterative computation under matched budgets. To this end, we present LoopMoE, a looped MoE language model that integrates sparse routing with iterative weight-shared computation through two designs. The first is IterAdaLN, which resolves weight-sharing symmetry via a modulation signal jointly conditioned on the iteration index and the per-token hidden state. The second is a capacity-balancing strategy that recovers the attention-to-FFN active parameter ratio of well-tuned non-looped references. Together, these designs enable the first strictly controlled, head-to-head evaluation of a looped MoE against a Vanilla MoE under identical total parameters, per-token FLOPs, and active sublayer ratios. At the 3B scale, LoopMoE outperforms the Vanilla MoE on 8 of 9 downstream benchmarks with an average improvement exceeding 1 point. At the 9B scale, LoopMoE continues to outperform the matched Vanilla MoE, indicating that the architectural gain persists at larger scale. Our work establishes a controlled synthesis of sparsity and recurrence, and suggests a promising direction for looped language models.

Cosmos 3: Omnimodal World Models for Physical AI

We introduce Cosmos 3, a family of omnimodal world models designed to jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture. By supporting highly flexible input-output configurations, Cosmos 3 seamlessly unifies critical modalities for Physical AI -- effectively subsuming vision-language models, video generators, world simulators, and world-action models into a single framework. Our evaluation demonstrates that Cosmos 3 establishes a new state-of-the-art across a diverse suite of understanding and generation tasks, demonstrating omnimodal world models as scalable, general-purpose backbones for embodied agents. Our post-trained Cosmos 3 models were ranked as the best open-source Text-to-Image and Image-to-Video models by Artificial Analysis, and the best policy model by RoboArena at the time the technical report was written. To accelerate open research and deployment in Physical AI, we make our code, model checkpoints, curated synthetic datasets, and evaluation benchmark available under the Linux Foundation's OpenMDW-1.1 https://openmdw.ai/license/1-1/ License at https://github.com/nvidia/cosmos}{github.com/nvidia/cosmos and https://huggingface.co/collections/nvidia/cosmos3 . The project website is available at https://research.nvidia.com/labs/cosmos-lab/cosmos3 .

PAVLM: Advancing Point Cloud based Affordance Understanding Via Vision-Language Model

Affordance understanding, the task of identifying actionable regions on 3D objects, plays a vital role in allowing robotic systems to engage with and operate within the physical world. Although Visual Language Models (VLMs) have excelled in high-level reasoning and long-horizon planning for robotic manipulation, they still fall short in grasping the nuanced physical properties required for effective human-robot interaction. In this paper, we introduce PAVLM (Point cloud Affordance Vision-Language Model), an innovative framework that utilizes the extensive multimodal knowledge embedded in pre-trained language models to enhance 3D affordance understanding of point cloud. PAVLM integrates a geometric-guided propagation module with hidden embeddings from large language models (LLMs) to enrich visual semantics. On the language side, we prompt Llama-3.1 models to generate refined context-aware text, augmenting the instructional input with deeper semantic cues. Experimental results on the 3D-AffordanceNet benchmark demonstrate that PAVLM outperforms baseline methods for both full and partial point clouds, particularly excelling in its generalization to novel open-world affordance tasks of 3D objects. For more information, visit our project site: pavlm-source.github.io.

CASE: Commonsense-Augmented Score with an Expanded Answer Space

LLMs have demonstrated impressive zero-shot performance on NLP tasks thanks to the knowledge they acquired in their training. In multiple-choice QA tasks, the LM probabilities are used as an imperfect measure of the plausibility of each answer choice. One of the major limitations of the basic score is that it treats all words as equally important. We propose CASE, a Commonsense-Augmented Score with an Expanded Answer Space. CASE addresses this limitation by assigning importance weights for individual words based on their semantic relations to other words in the input. The dynamic weighting approach outperforms basic LM scores, not only because it reduces noise from unimportant words, but also because it informs the model of implicit commonsense knowledge that may be useful for answering the question. We then also follow prior work in expanding the answer space by generating lexically-divergent answers that are conceptually-similar to the choices. When combined with answer space expansion, our method outperforms strong baselines on 5 commonsense benchmarks. We further show these two approaches are complementary and may be especially beneficial when using smaller LMs.

RestNet: Boosting Cross-Domain Few-Shot Segmentation with Residual Transformation Network

Cross-domain few-shot segmentation (CD-FSS) aims to achieve semantic segmentation in previously unseen domains with a limited number of annotated samples. Although existing CD-FSS models focus on cross-domain feature transformation, relying exclusively on inter-domain knowledge transfer may lead to the loss of critical intra-domain information. To this end, we propose a novel residual transformation network (RestNet) that facilitates knowledge transfer while retaining the intra-domain support-query feature information. Specifically, we propose a Semantic Enhanced Anchor Transform (SEAT) module that maps features to a stable domain-agnostic space using advanced semantics. Additionally, an Intra-domain Residual Enhancement (IRE) module is designed to maintain the intra-domain representation of the original discriminant space in the new space. We also propose a mask prediction strategy based on prototype fusion to help the model gradually learn how to segment. Our RestNet can transfer cross-domain knowledge from both inter-domain and intra-domain without requiring additional fine-tuning. Extensive experiments on ISIC, Chest X-ray, and FSS-1000 show that our RestNet achieves state-of-the-art performance. Our code will be available soon.

Semi-supervised Domain Adaptation via Prototype-based Multi-level Learning

In semi-supervised domain adaptation (SSDA), a few labeled target samples of each class help the model to transfer knowledge representation from the fully labeled source domain to the target domain. Many existing methods ignore the benefits of making full use of the labeled target samples from multi-level. To make better use of this additional data, we propose a novel Prototype-based Multi-level Learning (ProML) framework to better tap the potential of labeled target samples. To achieve intra-domain adaptation, we first introduce a pseudo-label aggregation based on the intra-domain optimal transport to help the model align the feature distribution of unlabeled target samples and the prototype. At the inter-domain level, we propose a cross-domain alignment loss to help the model use the target prototype for cross-domain knowledge transfer. We further propose a dual consistency based on prototype similarity and linear classifier to promote discriminative learning of compact target feature representation at the batch level. Extensive experiments on three datasets, including DomainNet, VisDA2017, and Office-Home demonstrate that our proposed method achieves state-of-the-art performance in SSDA.

Learning 6-DoF Task-oriented Grasp Detection via Implicit Estimation and Visual Affordance

Currently, task-oriented grasp detection approaches are mostly based on pixel-level affordance detection and semantic segmentation. These pixel-level approaches heavily rely on the accuracy of a 2D affordance mask, and the generated grasp candidates are restricted to a small workspace. To mitigate these limitations, we first construct a novel affordance-based grasp dataset and propose a 6-DoF task-oriented grasp detection framework, which takes the observed object point cloud as input and predicts diverse 6-DoF grasp poses for different tasks. Specifically, our implicit estimation network and visual affordance network in this framework could directly predict coarse grasp candidates, and corresponding 3D affordance heatmap for each potential task, respectively. Furthermore, the grasping scores from coarse grasps are combined with heatmap values to generate more accurate and finer candidates. Our proposed framework shows significant improvements compared to baselines for existing and novel objects on our simulation dataset. Although our framework is trained based on the simulated objects and environment, the final generated grasp candidates can be accurately and stably executed in real robot experiments when the object is randomly placed on a support surface.

Hard Sample Aware Noise Robust Learning for Histopathology Image Classification

Deep learning-based histopathology image classification is a key technique to help physicians in improving the accuracy and promptness of cancer diagnosis. However, the noisy labels are often inevitable in the complex manual annotation process, and thus mislead the training of the classification model. In this work, we introduce a novel hard sample aware noise robust learning method for histopathology image classification. To distinguish the informative hard samples from the harmful noisy ones, we build an easy/hard/noisy (EHN) detection model by using the sample training history. Then we integrate the EHN into a self-training architecture to lower the noise rate through gradually label correction. With the obtained almost clean dataset, we further propose a noise suppressing and hard enhancing (NSHE) scheme to train the noise robust model. Compared with the previous works, our method can save more clean samples and can be directly applied to the real-world noisy dataset scenario without using a clean subset. Experimental results demonstrate that the proposed scheme outperforms the current state-of-the-art methods in both the synthetic and real-world noisy datasets. The source code and data are available at https://github.com/bupt-ai-cz/HSA-NRL/.

Sample Prior Guided Robust Model Learning to Suppress Noisy Labels

Imperfect labels are ubiquitous in real-world datasets and seriously harm the model performance. Several recent effective methods for handling noisy labels have two key steps: 1) dividing samples into cleanly labeled and wrongly labeled sets by training loss, 2) using semi-supervised methods to generate pseudo-labels for samples in the wrongly labeled set. However, current methods always hurt the informative hard samples due to the similar loss distribution between the hard samples and the noisy ones. In this paper, we proposed PGDF (Prior Guided Denoising Framework), a novel framework to learn a deep model to suppress noise by generating the samples' prior knowledge, which is integrated into both dividing samples step and semi-supervised step. Our framework can save more informative hard clean samples into the cleanly labeled set. Besides, our framework also promotes the quality of pseudo-labels during the semi-supervised step by suppressing the noise in the current pseudo-labels generating scheme. To further enhance the hard samples, we reweight the samples in the cleanly labeled set during training. We evaluated our method using synthetic datasets based on CIFAR-10 and CIFAR-100, as well as on the real-world datasets WebVision and Clothing1M. The results demonstrate substantial improvements over state-of-the-art methods.