Vision-Language Navigation (VLN) enables robots to follow natural-language instructions in visually grounded environments, serving as a key capability for embodied robotic systems. Recent Vision-Language-Action (VLA) models have demonstrated strong navigation performance, but their high computational cost introduces latency that limits real-time deployment. We propose a training-free spatio-temporal vision token pruning framework tailored to VLA-based VLN. We apply spatial token selection to the current view, alongside spatio-temporal compression for historical memories, enabling efficient long-horizon inference while reducing redundant computation. Leveraging attention-based token importance and query-guided spatio-temporal filtering, the proposed approach preserves navigation-relevant information without retraining or modifying pretrained models, allowing plug-and-play integration into existing VLA systems. Through experiments on standard VLN benchmarks, we confirm that our method significantly outperforms existing pruning strategies. It successfully preserves superior navigation accuracy under extreme pruning scenarios, all while maintaining the highly competitive inference efficiency. Real-world deployment on a Unitree Go2 quadruped robot further validates reliable and low-latency instruction-following navigation under practical robotic constraints. We hope this work helps bridge the gap between large-scale multimodal modeling and efficient, real-time embodied deployment in robotic navigation systems.

The development of radiology foundation models (RFMs) is hindered by a reliance on brute-force scaling. Existing approaches often directly translate methods for natural images, which prioritize scale over precision and hence lead to brittle and expensive models in clinical practice. To address this, we present a resource-efficient pre-training framework, GreenRFM, that achieves state-of-the-art performance. Our framework ensures robust generalization across diverse patient populations and imaging protocols, reducing computational requirements by orders of magnitude while surpassing complex, parameter-heavy models. These capabilities stem from principled supervision design that aims to maximally utilize supervisory signals via More distilled, Ubiquitous, Semantic-enforcing, and Task-aligning (MUST) supervision, rather than simply piling up the quantity of training data. We offer two GreenRFM configurations: (i) a performant model that establishes a new state-of-the-art using a single 24GB GPU within 24 hours, and (ii) a lightweight model that matches existing benchmarks with 6GB VRAM in 4 hours. We conduct extensive experiments using over 200,000 images from four institutions and of two modalities. GreenRFMs achieve superior performances on chest and abdominal CT datasets, regardless of public or private benchmark, surpassing a range of baseline models. In addition, the results on internal musculoskeletal MRI images show that the same supervision principles transfer between different modalities. Our performance and efficiency challenge the ``scale is all you need'' dogma and democratize the equitable development of state-of-the-art RFMs for clinicians even on a laptop.

Vision-language models encode continuous geometry that their text pathway fails to express: a 6,000-parameter linear probe extracts hand joint angles at 6.1 degrees MAE from frozen features, while the best text output achieves only 20.0 degrees -- a 3.3x bottleneck. LoRA fine-tuning (r=16, 2,000 images) narrows this gap to 6.5 degrees, providing evidence for a pathway-training deficit rather than a representational one. Training objective determines accuracy more than architecture: five encoders spanning self-supervised, contrastive, and hybrid paradigms converge to statistically equivalent accuracy (R^2 approximately 0.55, TOST-equivalent at delta=0.03) despite sharing as little as CKA=0.41 representational similarity -- functional convergence without representational convergence. Autoregressive generation damages geometric fidelity, but the damage originates in the generation process, not in language alignment: Qwen2.5-VL's LLM layers actually improve probe accuracy over its raw vision encoder. Layer-wise analysis reveals a universal mid-network accuracy peak across all architectures, with attention heads in layers 18-22 carrying disproportionate geometric signal. These findings enable a single frozen backbone to function as a multi-task geometric sensor through lightweight probes, without fine-tuning or text generation.

Autoregressive (AR) language models rely on causal tokenization, but extending this paradigm to vision remains non-trivial. Current visual tokenizers either flatten 2D patches into non-causal sequences or enforce heuristic orderings that misalign with the "next-token prediction" pattern. Recent diffusion autoencoders similarly fall short: conditioning the decoder on all tokens lacks causality, while applying nested dropout mechanism introduces imbalance. To address these challenges, we present CaTok, a 1D causal image tokenizer with a MeanFlow decoder. By selecting tokens over time intervals and binding them to the MeanFlow objective, as illustrated in Fig. 1, CaTok learns causal 1D representations that support both fast one-step generation and high-fidelity multi-step sampling, while naturally capturing diverse visual concepts across token intervals. To further stabilize and accelerate training, we propose a straightforward regularization REPA-A, which aligns encoder features with Vision Foundation Models (VFMs). Experiments demonstrate that CaTok achieves state-of-the-art results on ImageNet reconstruction, reaching 0.75 FID, 22.53 PSNR and 0.674 SSIM with fewer training epochs, and the AR model attains performance comparable to leading approaches.

Situated reasoning often relies on active exploration, yet in many real-world scenarios such exploration is infeasible due to physical constraints of robots or safety concerns of visually impaired users. Given only a limited observation, can an agent mentally simulate a future trajectory toward a target situation and answer spatial what-if questions? We introduce WanderDream, the first large-scale dataset designed for the emulative simulation of mental exploration, enabling models to reason without active exploration. WanderDream-Gen comprises 15.8K panoramic videos across 1,088 real scenes from HM3D, ScanNet++, and real-world captures, depicting imagined trajectories from current viewpoints to target situations. WanderDream-QA contains 158K question-answer pairs, covering starting states, paths, and end states along each trajectory to comprehensively evaluate exploration-based reasoning. Extensive experiments with world models and MLLMs demonstrate (1) that mental exploration is essential for situated reasoning, (2) that world models achieve compelling performance on WanderDream-Gen, (3) that imagination substantially facilitates reasoning on WanderDream-QA, and (4) that WanderDream data exhibit remarkable transferability to real-world scenarios. The source code and all data will be released.

Streaming TTS that receives streaming text is essential for interactive systems, yet this scheme faces two major challenges: unnatural prosody due to missing lookahead and long-form collapse due to unbounded context. We propose a prosodic-boundary-aware post-training strategy, adapting a pretrained LLM-based TTS model using weakly time-aligned data. Specifically, the model is adapted to learn early stopping at specified content boundaries when provided with limited future text. During inference, a sliding-window prompt carries forward previous text and speech tokens, ensuring bounded context and seamless concatenation. Evaluations show our method outperforms CosyVoice-Style interleaved baseline in both short and long-form scenarios. In long-text synthesis, especially, it achieves a 66.2% absolute reduction in word error rate (from 71.0% to 4.8%) and increases speaker and emotion similarity by 16.1% and 1.5% relatively, offering a robust solution for streaming TTS with incremental text.

We propose a Correlation-Complexity Map as a practical diagnostic tool for determining when real-world data distributions are structurally aligned with IQP-type quantum generative models. Characterized by two complementary indicators: (i) a Quantum Correlation-Likeness Indicator (QCLI), computed from the dataset's correlation-order (Walsh-Hadamard/Fourier) power spectrum aggregated by interaction order and quantified via Jensen-Shannon divergence from an i.i.d. binomial reference; and (ii) a Classical Correlation-Complexity Indicator (CCI), defined as the fraction of total correlation not captured by the optimal Chow-Liu tree approximation, normalized by total correlation. We provide theoretical support by relating QCLI to a support-mismatch mechanism, for fixed-architecture IQP families trained with an MMD objective, higher QCLI implies a smaller irreducible approximation floor. Using the map, we identify the classical turbulence data as both IQP-compatible and classically complex (high QCLI/high CCI). Guided by this placement, we use an invertible float-to-bitstring representation and a latent-parameter adaptation scheme that reuses a compact IQP circuit over a temporal sequence by learning and interpolating a low-dimensional latent trajectory. In comparative evaluations against classical models such as Restricted Boltzmann Machine (RBM) and Deep Convolutional Generative Adversarial Networks (DCGAN), the IQP approach achieves competitive distributional alignment while using substantially fewer training snapshots and a small latent block, supporting the use of QCLI/CCI as practical indicators for locating IQP-aligned domains and advancing generative quantum utility.

Research in AI using Large-Language Models (LLMs) is rapidly evolving, and the comparison of their performance with human reasoning has become a key concern. Prior studies have indicated that LLMs and humans share similar biases, such as dismissing logically valid inferences that contradict common beliefs. However, criticizing LLMs for these biases might be unfair, considering our reasoning not only involves formal deduction but also abduction, which draws tentative conclusions from limited information. Abduction can be regarded as the inverse form of syllogism in its basic structure, that is, a process of drawing a minor premise from a major premise and conclusion. This paper explores the accuracy of LLMs in abductive reasoning by converting a syllogistic dataset into one suitable for abduction. It aims to investigate whether the state-of-the-art LLMs exhibit biases in abduction and to identify potential areas for improvement, emphasizing the importance of contextualized reasoning beyond formal deduction. This investigation is vital for advancing the understanding and application of LLMs in complex reasoning tasks, offering insights into bridging the gap between machine and human cognition.

Interactive segmentation enables clinicians to guide annotation, but existing zero-shot models like nnInteractive fail to consistently reach expert-level performance across diverse medical imaging tasks. Because annotation campaigns produce a growing stream of task-specific labelled data, online adaptation of the segmentation model is a natural complement to zero-shot inference. We propose CLoPA, a continual adaptation strategy that tunes a small fraction of nnInteractive's parameters on the annotation cache, triggered by lightweight episode scheduling. CLoPA requires no new parameters or changes to the inference pipeline, and operates entirely within the existing annotation workflow. Across eight Medical Segmentation Decathlon tasks spanning diverse anatomical targets and imaging characteristics, CLoPA rapidly elevates performance to expert-level, even for tasks where nnInteractive previously failed, with the majority of gains realised after a single training episode. We show that the benefits of tuning different parameter groups depends on task characteristics and data regimes. Also, that for targets with complex geometries (e.g., hepatic vessels), instance normalisation and low-level feature tuning saturates, suggesting a need for deeper feature-representation alignment in the most challenging scenarios.

Large language models (LLMs) have recently reshaped Automated Essay Scoring (AES), yet prior studies typically examine individual techniques in isolation, limiting understanding of their relative merits for English as a Second Language (L2) writing. To bridge this gap, we presents a comprehensive comparison of major LLM-based AES paradigms on IELTS Writing Task~2. On this unified benchmark, we evaluate four approaches: (i) encoder-based classification fine-tuning, (ii) zero- and few-shot prompting, (iii) instruction tuning and Retrieval-Augmented Generation (RAG), and (iv) Supervised Fine-Tuning combined with Direct Preference Optimization (DPO) and RAG. Our results reveal clear accuracy-cost-robustness trade-offs across methods, the best configuration, integrating k-SFT and RAG, achieves the strongest overall results with F1-Score 93%. This study offers the first unified empirical comparison of modern LLM-based AES strategies for English L2, promising potential in auto-grading writing tasks. Code is public at https://github.com/MinhNguyenDS/LLM_AES-EnL2

Monitoring fish growth behavior provides relevant information about fish health in aquaculture and home aquariums. Yet, monitoring fish sizes poses different challenges, as fish are small and subject to strong refractive distortions in aquarium environments. Image-based measurement offers a practical, non-invasive alternative that allows frequent monitoring without disturbing the fish. In this paper, we propose a non-invasive refraction-aware stereo vision method to estimate fish length in aquariums. Our approach uses a YOLOv11-Pose network to detect fish and predict anatomical keypoints on the fish in each stereo image. A refraction-aware epipolar constraint accounting for the air-glass-water interfaces enables robust matching, and unreliable detections are removed using a learned quality score. A subsequent refraction-aware 3D triangulation recovers 3D keypoints, from which fish length is measured. We validate our approach on a new stereo dataset of endangered Sulawesi ricefish captured under aquarium-like conditions and demonstrate that filtering low-quality detections is essential for accurate length estimation. The proposed system offers a simple and practical solution for non-invasive growth monitoring and can be easily applied in home aquariums.

As large language models (LLMs) advance in linguistic competence, their reasoning abilities are gaining increasing attention. In humans, reasoning often performs well in domain specific settings, particularly in normative rather than purely formal contexts. Although prior studies have compared LLM and human reasoning, the domain specificity of LLM reasoning remains underexplored. In this study, we introduce a new Wason Selection Task dataset that explicitly encodes deontic modality to systematically distinguish deontic from descriptive conditionals, and use it to examine LLMs' conditional reasoning under deontic rules. We further analyze whether observed error patterns are better explained by confirmation bias (a tendency to seek rule-supporting evidence) or by matching bias (a tendency to ignore negation and select items that lexically match elements of the rule). Results show that, like humans, LLMs reason better with deontic rules and display matching-bias-like errors. Together, these findings suggest that the performance of LLMs varies systematically across rule types and that their error patterns can parallel well-known human biases in this paradigm.

Recent advances in diffusion-based video generation have achieved remarkable visual realism but still struggle to obey basic physical laws such as gravity, inertia, and collision. Generated objects often move inconsistently across frames, exhibit implausible dynamics, or violate physical constraints, limiting the realism and reliability of AI-generated videos. We address this gap by introducing Physical Simulator In-the-loop Video Generation (PSIVG), a novel framework that integrates a physical simulator into the video diffusion process. Starting from a template video generated by a pre-trained diffusion model, PSIVG reconstructs the 4D scene and foreground object meshes, initializes them within a physical simulator, and generates physically consistent trajectories. These simulated trajectories are then used to guide the video generator toward spatio-temporally physically coherent motion. To further improve texture consistency during object movement, we propose a Test-Time Texture Consistency Optimization (TTCO) technique that adapts text and feature embeddings based on pixel correspondences from the simulator. Comprehensive experiments demonstrate that PSIVG produces videos that better adhere to real-world physics while preserving visual quality and diversity. Project Page: https://vcai.mpi-inf.mpg.de/projects/PSIVG/

Vision Transformers must resolve figure-ground organization by choosing between completions driven by local geometric evidence and those favored by global organizational priors, giving rise to a characteristic perceptual ambiguity. We aim to locate where the canonical Gestalt prior convexity is realized within the internal components of BEiT. Using a controlled perceptual conflict based on synthetic shapes of darts, we systematically mask regions that equally admit either a concave completion or a convex completion. We show that BEiT reliably favors convex completion under this competition. Projecting internal activations into the model's discrete visual codebook space via logit attribution reveals that this preference is governed by identifiable functional units within transformer substructures. Specifically, we find that figure-ground organization is ambiguous through early and intermediate layers and resolves abruptly in later layers. By decomposing the direct effect of attention heads, we identify head L0H9 acting as an early seed, introducing a weak bias toward convexity. Downscaling this single attention head shifts the distributional mass of the perceptual conflict across a continuous decision boundary, allowing concave evidence to guide completion.

ROScopter is a lean multirotor autopilot built for researchers. ROScopter seeks to accelerate simulation and hardware testing of research code with an architecture that is both easy to understand and simple to modify. ROScopter is designed to interface with ROSflight 2.0 and runs entirely on an onboard flight computer, leveraging the features of ROS 2 to improve modularity. This work describes the architecture of ROScopter and how it can be used to test application code in both simulated and hardware environments. Hardware results of the default ROScopter behavior are presented, showing that ROScopter achieves similar performance to another state-of-the-art autopilot for basic waypoint-following maneuvers, but with a significantly reduced and more modular code-base.

A critical bottleneck in robot learning is the scarcity of task-labeled, segmented training data, despite the abundance of large-scale robotic datasets recorded as long, continuous interaction logs. Existing datasets contain vast amounts of diverse behaviors, yet remain structurally incompatible with modern learning frameworks that require cleanly segmented, task-specific trajectories. We address this data utilization crisis by formalizing robotic sequence retrieval: the task of extracting reusable, task-centric segments from unlabeled logs using only a few reference examples. We introduce ROSER, a lightweight few-shot retrieval framework that learns task-agnostic metric spaces over temporal windows, enabling accurate retrieval with as few as 3-5 demonstrations, without any task-specific training required. To validate our approach, we establish comprehensive evaluation protocols and benchmark ROSER against classical alignment methods, learned embeddings, and language model baselines across three large-scale datasets (e.g., LIBERO, DROID, and nuScenes). Our experiments demonstrate that ROSER consistently outperforms all prior methods in both accuracy and efficiency, achieving sub-millisecond per-match inference while maintaining superior distributional alignment. By reframing data curation as few-shot retrieval, ROSER provides a practical pathway to unlock underutilized robotic datasets, fundamentally improving data availability for robot learning.

AI agents are increasingly developed and evaluated on benchmarks relevant to human work, yet it remains unclear how representative these benchmarking efforts are of the labor market as a whole. In this work, we systematically study the relationship between agent development efforts and the distribution of real-world human work by mapping benchmark instances to work domains and skills. We first analyze 43 benchmarks and 72,342 tasks, measuring their alignment with human employment and capital allocation across all 1,016 real-world occupations in the U.S. labor market. We reveal substantial mismatches between agent development that tends to be programming-centric, and the categories in which human labor and economic value are concentrated. Within work areas that agents currently target, we further characterize current agent utility by measuring their autonomy levels, providing practical guidance for agent interaction strategies across work scenarios. Building on these findings, we propose three measurable principles for designing benchmarks that better capture socially important and technically challenging forms of work: coverage, realism, and granular evaluation.

Mobile Agents can autonomously execute user instructions, which requires hybrid-capabilities reasoning, including screen summary, subtask planning, action decision and action function. However, existing agents struggle to achieve both decoupled enhancement and balanced integration of these capabilities. To address these challenges, we propose Channel-of-Mobile-Experts (CoME), a novel agent architecture consisting of four distinct experts, each aligned with a specific reasoning stage, CoME activates the corresponding expert to generate output tokens in each reasoning stage via output-oriented activation. To empower CoME with hybrid-capabilities reasoning, we introduce a progressive training strategy: Expert-FT enables decoupling and enhancement of different experts' capability; Router-FT aligns expert activation with the different reasoning stage; CoT-FT facilitates seamless collaboration and balanced optimization across multiple capabilities. To mitigate error propagation in hybrid-capabilities reasoning, we propose InfoGain-Driven DPO (Info-DPO), which uses information gain to evaluate the contribution of each intermediate step, thereby guiding CoME toward more informative reasoning. Comprehensive experiments show that CoME outperforms dense mobile agents and MoE methods on both AITZ and AMEX datasets.

We present a geometry-aware method for heterogeneous single-particle cryogenic electron microscopy (cryo-EM) reconstruction that predicts atomic backbone conformations. To incorporate protein-structure priors, we represent the backbone as a graph and use a graph neural network (GNN) autodecoder that maps per-image latent variables to 3D displacements of a template conformation. The objective combines a data-discrepancy term based on a differentiable cryo-EM forward model with geometric regularization, and it supports unknown orientations via ellipsoidal support lifting (ESL) pose estimation. On synthetic datasets derived from molecular dynamics trajectories, the proposed GNN achieves higher accuracy compared to a multilayer perceptron (MLP) of comparable size, highlighting the benefits of a geometry-informed inductive bias.

Recent advances in large language models (LLMs) have opened new avenues for accelerating scientific research. While models are increasingly capable of assisting with routine tasks, their ability to contribute to novel, expert-level mathematical discovery is less understood. We present a collection of case studies demonstrating how researchers have successfully collaborated with advanced AI models, specifically Google's Gemini-based models (in particular Gemini Deep Think and its advanced variants), to solve open problems, refute conjectures, and generate new proofs across diverse areas in theoretical computer science, as well as other areas such as economics, optimization, and physics. Based on these experiences, we extract common techniques for effective human-AI collaboration in theoretical research, such as iterative refinement, problem decomposition, and cross-disciplinary knowledge transfer. While the majority of our results stem from this interactive, conversational methodology, we also highlight specific instances that push beyond standard chat interfaces. These include deploying the model as a rigorous adversarial reviewer to detect subtle flaws in existing proofs, and embedding it within a "neuro-symbolic" loop that autonomously writes and executes code to verify complex derivations. Together, these examples highlight the potential of AI not just as a tool for automation, but as a versatile, genuine partner in the creative process of scientific discovery.

Large language models (LLMs) have demonstrated strong reasoning capabilities on math and coding, but frequently fail on symbolic classical planning tasks. Our studies, as well as prior work, show that LLM-generated plans routinely violate domain constraints given in their instructions (e.g., walking through walls). To address this failure, we propose iteratively augmenting instructions with Localized In-Context Learning (L-ICL) demonstrations: targeted corrections for specific failing steps. Specifically, L-ICL identifies the first constraint violation in a trace and injects a minimal input-output example giving the correct behavior for the failing step. Our proposed technique of L-ICL is much effective than explicit instructions or traditional ICL, which adds complete problem-solving trajectories, and many other baselines. For example, on an 8x8 gridworld, L-ICL produces valid plans 89% of the time with only 60 training examples, compared to 59% for the best baseline, an increase of 30%. L-ICL also shows dramatic improvements in other domains (gridworld navigation, mazes, Sokoban, and BlocksWorld), and on several LLM architectures.

We present PepEDiff, a novel peptide binder generator that designs binding sequences given a target receptor protein sequence and its pocket residues. Peptide binder generation is critical in therapeutic and biochemical applications, yet many existing methods rely heavily on intermediate structure prediction, adding complexity and limiting sequence diversity. Our approach departs from this paradigm by generating binder sequences directly in a continuous latent space derived from a pretrained protein embedding model, without relying on predicted structures, thereby improving structural and sequence diversity. To encourage the model to capture binding-relevant features rather than memorizing known sequences, we perform latent-space exploration and diffusion-based sampling, enabling the generation of peptides beyond the limited distribution of known binders. This zero-shot generative strategy leverages the global protein embedding manifold as a semantic prior, allowing the model to propose novel peptide sequences in previously unseen regions of the protein space. We evaluate PepEDiff on TIGIT, a challenging target with a large, flat protein-protein interaction interface that lacks a druggable pocket. Despite its simplicity, our method outperforms state-of-the-art approaches across benchmark tests and in the TIGIT case study, demonstrating its potential as a general, structure-free framework for zero-shot peptide binder design. The code for this research is available at GitHub: https://github.com/LabJunBMI/PepEDiff-An-Peptide-binder-Embedding-Diffusion-Model

This paper presents InsSo3D, an accurate and efficient method for large-scale 3D Simultaneous Localisation and Mapping (SLAM) using a 3D Sonar and an Inertial Navigation System (INS). Unlike traditional sonar, which produces 2D images containing range and azimuth information but lacks elevation information, 3D Sonar produces a 3D point cloud, which therefore does not suffer from elevation ambiguity. We introduce a robust and modern SLAM framework adapted to the 3D Sonar data using INS as prior, detecting loop closure and performing pose graph optimisation. We evaluated InsSo3D performance inside a test tank with access to ground truth data and in an outdoor flooded quarry. Comparisons to reference trajectories and maps obtained from an underwater motion tracking system and visual Structure From Motion (SFM) demonstrate that InsSo3D efficiently corrects odometry drift. The average trajectory error is below 21cm during a 50-minute-long mission, producing a map of 10m by 20m with a 9cm average reconstruction error, enabling safe inspection of natural or artificial underwater structures even in murky water conditions.

Visual challenges in underwater environments significantly hinder the accuracy of vision-based localisation and the high-fidelity dense reconstruction. In this paper, we propose VISO, a robust underwater SLAM system that fuses a stereo camera, an inertial measurement unit (IMU), and a 3D sonar to achieve accurate 6-DoF localisation and enable efficient dense 3D reconstruction with high photometric fidelity. We introduce a coarse-to-fine online calibration approach for extrinsic parameters estimation between the 3D sonar and the camera. Additionally, a photometric rendering strategy is proposed for the 3D sonar point cloud to enrich the sonar map with visual information. Extensive experiments in a laboratory tank and an open lake demonstrate that VISO surpasses current state-of-the-art underwater and visual-based SLAM algorithms in terms of localisation robustness and accuracy, while also exhibiting real-time dense 3D reconstruction performance comparable to the offline dense mapping method.

4D spatial intelligence involves perceiving and processing how objects move or change over time. Humans naturally possess 4D spatial intelligence, supporting a broad spectrum of spatial reasoning abilities. To what extent can Multimodal Large Language Models (MLLMs) achieve human-level 4D spatial intelligence? In this work, we present Spatial4D-Bench, a versatile 4D spatial intelligence benchmark designed to comprehensively assess the 4D spatial reasoning abilities of MLLMs. Unlike existing spatial intelligence benchmarks that are often small-scale or limited in diversity, Spatial4D-Bench provides a large-scale, multi-task evaluation benchmark consisting of ~40,000 question-answer pairs covering 18 well-defined tasks. We systematically organize these tasks into six cognitive categories: object understanding, scene understanding, spatial relationship understanding, spatiotemporal relationship understanding, spatial reasoning and spatiotemporal reasoning. Spatial4D-Bench thereby offers a structured and comprehensive benchmark for evaluating the spatial cognition abilities of MLLMs, covering a broad spectrum of tasks that parallel the versatility of human spatial intelligence. We benchmark various state-of-the-art open-source and proprietary MLLMs on Spatial4D-Bench and reveal their substantial limitations in a wide variety of 4D spatial reasoning aspects, such as route plan, action recognition, and physical plausibility reasoning. We hope that the findings provided in this work offer valuable insights to the community and that our benchmark can facilitate the development of more capable MLLMs toward human-level 4D spatial intelligence. More resources can be found on our project page.

Vision-language models (VLMs) are commonly trained by directly inserting image tokens from a pretrained vision encoder into the text stream of a language model. This allows text and image information to fully attend to one another within the model, but becomes rapidly costly for long multi-image conversations or streaming video applications, both in terms of memory and compute. VLMs leveraging cross-attention (CA) are an efficient alternative to token insertion as image tokens are not added to the KV cache. Despite being introduced early on, multimodal CA models are scarce in the current VLM literature and often underperform their token insertion counterparts. In this work, we reinvestigate the effectiveness of cross-attention for vision-language modeling: (i) We analyze the core differences between the cross-attention and self-attention mechanisms, (ii) we train cross-attention VLMs both from a text-only LLM and by adapting a pretrained insertion-based VLM, showing that simple cross-attention is far more competitive with token insertion than previously reported, and (iii) we demonstrate the practical advantages of cross-attention on real-time video captioning, where it naturally maintains low latency and near-constant memory cost. For samples and code, please see our project page at https://kyutai.org/casa .

The deployment of complex soft robots in multiphysics environments requires advanced simulation frameworks that not only capture interactions between different types of material, but also translate accurately to real-world performance. Soft robots pose unique modeling challenges due to their large nonlinear deformations, material incompressibility, and contact interactions, which complicate both numerical stability and physical accuracy. Despite recent progress, robotic simulators often struggle with modeling such phenomena in a scalable and application-relevant manner. We present SORS (Soft Over Rigid Simulator), a versatile, high-fidelity simulator designed to handle these complexities for soft robot applications. Our energy-based framework, built on the finite element method, allows modular extensions, enabling the inclusion of custom-designed material and actuation models. To ensure physically consistent contact handling, we integrate a constrained nonlinear optimization based on sequential quadratic programming, allowing for stable and accurate modeling of contact phenomena. We validate our simulator through a diverse set of real-world experiments, which include cantilever deflection, pressure-actuation of a soft robotic arm, and contact interactions from the PokeFlex dataset. In addition, we showcase the potential of our framework for control optimization of a soft robotic leg. These tests confirm that our simulator can capture both fundamental material behavior and complex actuation dynamics with high physical fidelity. By bridging the sim-to-real gap in these challenging domains, our approach provides a validated tool for prototyping next-generation soft robots, filling the gap of extensibility, fidelity, and usability in the soft robotic ecosystem.

The quality of the latent space in visual tokenizers (e.g., VAEs) is crucial for modern generative models. However, the standard reconstruction-based training paradigm produces a latent space that is biased towards low-level information, leading to a foundation flaw: better pixel-level accuracy does not lead to higher-quality generation. This implies that pouring extensive compute into visual tokenizer pre-training translates poorly to improved performance in generation. We identify this as the ``pre-training scaling problem`` and suggest a necessary shift: to be effective for generation, a latent space must concisely represent high-level semantics. We present VTP, a unified visual tokenizer pre-training framework, pioneering the joint optimization of image-text contrastive, self-supervised, and reconstruction losses. Our large-scale study reveals two principal findings: (1) understanding is a key driver of generation, and (2) much better scaling properties, where generative performance scales effectively with compute, parameters, and data allocated to the pretraining of the visual tokenizer. After large-scale pre-training, our tokenizer delivers a competitive profile (78.2 zero-shot accuracy and 0.36 rFID on ImageNet) and 4.1 times faster convergence on generation compared to advanced distillation methods. More importantly, it scales effectively: without modifying standard DiT training specs, solely investing more FLOPS in pretraining VTP achieves 65.8\% FID improvement in downstream generation, while conventional autoencoder stagnates very early at 1/10 FLOPS. Our pre-trained models are available at https://github.com/MiniMax-AI/VTP.

Reinforcement learning (RL) approaches based on Markov Decision Processes (MDPs) are predominantly applied in the robot joint space, often relying on limited task-specific information and partial awareness of the 3D environment. In contrast, episodic RL has demonstrated advantages over traditional MDP-based methods in terms of trajectory consistency, task awareness, and overall performance in complex robotic tasks. Moreover, traditional step-wise and episodic RL methods often neglect the contact-rich information inherent in task-space manipulation, especially considering the contact-safety and robustness. In this work, contact-rich manipulation tasks are tackled using a task-space, energy-safe framework, where reliable and safe task-space trajectories are generated through the combination of Proximal Policy Optimization (PPO) and movement primitives. Furthermore, an energy-aware Cartesian Impedance Controller objective is incorporated within the proposed framework to ensure safe interactions between the robot and the environment. Our experimental results demonstrate that the proposed framework outperforms existing methods in handling tasks on various types of surfaces in 3D environments, achieving high success rates as well as smooth trajectories and energy-safe interactions.

We present a novel framework for automated interior design that combines large language models (LLMs) with grid-based integer programming to jointly optimize room layout and furniture placement. Given a textual prompt, the LLM-driven agent workflow extracts structured design constraints related to room configurations and furniture arrangements. These constraints are encoded into a unified grid-based representation inspired by ``Modulor". Our formulation accounts for key design requirements, including corridor connectivity, room accessibility, spatial exclusivity, and user-specified preferences. To improve computational efficiency, we adopt a coarse-to-fine optimization strategy that begins with a low-resolution grid to solve a simplified problem and guides the solution at the full resolution. Experimental results across diverse scenarios demonstrate that our joint optimization approach significantly outperforms existing two-stage design pipelines in solution quality, and achieves notable computational efficiency through the coarse-to-fine strategy.