ICLR 2026 论文列表

The dominant paradigm for training Large Vision-Language Models (LVLMs) in navigation relies on imitating expert trajectories. This approach reduces the complex navigation task to a sequence-to-sequence replication of a single correct path, fundamentally limiting the agent's ability to explore and generalize. In this work, we argue for and introduce a new paradigm: a shift from Path Imitation to Decision Understanding. The goal of this paradigm is to build agents that do not just follow, but truly understand how to navigate. We materialize this through two core contributions: first, we introduce Compass-Data-22k, a novel 22k-trajectory dataset.Its Reinforcement Fine-Tuning (RFT) subset provides a panoramic view of the decision landscape by annotating all feasible actions with A* geodesic distances. Second, we design a novel gap-aware hybrid reward function that dynamically adapts its feedback to decision certainty, shifting between decisive signals for optimal actions and nuanced scores to encourage exploration. Integrated into an SFT-then-RFT recipe, our CompassNav agent is trained not to memorize static routes, but to develop an internal 'compass' that constantly intuits the direction to the goal by evaluating the relative quality of all possible moves. This approach enables our 7B agent to set a new state-of-the-art on navigation benchmarks, outperforming even larger proprietary models, and achieve robust real-world goal navigation on a physical robot.

Vision-based 3D human motion capture from videos remains a challenge in computer vision. Traditional 3D pose estimation approaches often ignore the temporal consistency between frames, causing implausible and jittery motion. The emerging field of kinematics-based 3D motion capture addresses these issues by estimating the temporal transitioning between poses instead. A major drawback in current kinematics approaches is their reliance on Euler angles. Despite their simplicity, Euler angles suffer from discontinuity that leads to unstable motion reconstructions, especially in online settings where trajectory refinement is unavailable. Contrarily, quaternions have no discontinuity and can produce continuous transitions between poses. In this paper, we propose QuaMo, a novel Quaternion Motions method using quaternion differential equations (QDE) for human kinematics capture. We utilize the state-space model, an effective system for describing real-time kinematics estimations, with quaternion state and the QDE describing quaternion velocity. The corresponding angular acceleration are computed from a meta-PD controller with a novel acceleration enhancement that adaptively regulates the control signals as the human quickly change to new pose. Unlike previous work, our QDE is solved under the quaternion geometric constraints that results in more accurate estimations. Experimental results show that our novel formulation of the QDE with acceleration enhancement accurately estimates 3D human kinematics with no discontinuity and minimal implausible artifact. QuaMo outperforms comparable state-of-the-art methods on multiple datasets, namely Human3.6M, Fit3D, SportsPose and a subset of AIST. The code is available at https://github.com/cuongle1206/QuaMo

Hallucination has recently garnered significant research attention in Large Vision-Language Models (LVLMs). Direct Preference Optimization (DPO) aims to learn directly from the corrected preferences provided by humans, thereby addressing the hallucination issue. Despite its success, this paradigm has yet to specifically target two critical causes of visual hallucination: the perceptual bottleneck in attended regions and insufficient Visual Robustness against image degradation. Furthermore, existing preference pairs are constructed by directly editing textual outputs without visual signals, and their off-policy nature limits effectiveness in guiding model learning. To address these challenges, we propose Perceptual Processing Direct Preference Optimization (P$^2$-DPO), a novel training paradigm where the model generates and learns from its own preference pairs, thereby directly addressing the identified visual bottlenecks while inherently avoiding the issues of vision-agnostic and off-policy data. It introduces: (1) an on-policy preference pair construction method targeting Focus-and-Enhance perception and Visual Robustness, and (2) a well-designed Calibration Loss to precisely align visual signals with the causal generation of text. Experimental results demonstrate that with a comparable amount of training data and cost, P$^2$-DPO outperforms even state-of-the-art methods that rely on costly human feedback on benchmarks such as POPE and MMHal-Bench. Furthermore, evaluations on Attention Region Fidelity (ARF) and image degradation scenarios validate the effectiveness of P$^2$-DPO in addressing perceptual bottlenecks in attended regions and improving Visual Robustness against degraded inputs.

Neural network pruning is a promising technique to mitigate the excessive computational and memory requirements of large language models (LLMs). Despite its promise, however, progress in this area has diminished, as conventional methods are seemingly unable to surpass moderate sparsity levels (50-60\%) without severely degrading model accuracy. This work breaks through the current impasse, presenting a principled and effective method called $ \text{Elsa}$, which achieves extreme sparsity levels of up to 90\% while retaining high model fidelity. This is done by identifying several limitations in current practice, all of which can be traced back to their reliance on a surrogate objective formulation. $ \text{Elsa}$ tackles this issue directly and effectively via standard and well-established constrained optimization techniques based on ADMM. Our extensive experiments across a wide range of models and scales show that $ \text{Elsa}$ achieves substantial improvements over existing methods; e.g., it achieves 7.8$ \times$ less perplexity than the best existing method on LLaMA-2-7B at 90\% sparsity. Moreover, we show that $ \text{Elsa}$ remains stable even at extreme sparsity (e.g., 95\%), yielding up to $\times$3.98 inference speedup and $\times$7.80 memory compression over its dense counterpart. We also present $ \text{Elsa}_ {-L}$, a quantized variant that scales to extremely large models (27B), and establish its theoretical convergence guarantees. These results highlight meaningful progress in advancing the frontier of LLM sparsity, while promising that significant opportunities for further advancement may remain in directions that have so far attracted limited exploration.

Vision-language (VL) models are increasingly deployed in non-stationary settings, yet under sequential adaptation they often preserve primitive recognition while losing compositional structure, especially with tight rehearsal budgets and no task IDs. We address this gap by asking how a continual VL system can maintain structurally dependable behaviour while safeguarding zero-shot performance. We introduce Compo-ReAlign, a structure-first recipe built around three components: a reversible composer that maps primitive embeddings to compositions by design, a multi-positive InfoNCE that jointly aligns textual and composed views of the same target, and a spectral trust region that clips updates when alignment sensitivity inflates. Across compositional DIL and multi-domain MTIL retrieval, Compo-ReAlign sets a new state of the art, improves over the strongest prior by +2.4 R@1, and reduces forgetting by 40%. We provide a compact, reversible alignment head with geometry-aware training for compositionally robust VL continual learning.

Large Language Models (LLMs) excel in natural language processing (NLP) tasks but often generate false or misleading information, known as hallucinations, raising reliability concerns in high-stakes applications. To provide statistical guarantees on the factuality of LLM outputs, conformal prediction based techniques have been proposed. Despite their strong theoretical guarantees, they rely heavily on the exchangeability assumption between calibration and test data, which is frequently violated in real-world scenarios with dynamic covariate shifts. To overcome this limitation, we introduce **CoFact**, a conformal prediction framework that uses online density ratio estimation to adaptively reweigh calibration data, ensuring alignment with evolving test distributions. With this approach, CoFact bypasses the exchangeability requirement and provides robust factuality guarantees under non-stationary conditions. To theoretically justify CoFact, we establish an upper bound on the gap between the actual hallucination rate and the target level $\alpha$, demonstrating that the bound asymptotically approaches zero as the number of rounds and calibration samples increase. Empirically, CoFact is evaluated on MedLFQA, WikiData, and the newly introduced **WildChat+** dataset, which captures real-world covariate shifts through user-generated prompts. Results demonstrate that CoFact consistently outperforms existing methods, maintaining reliability even under dynamic conditions.

We introduce Dynamic Nested Depth (DND), a novel method that improves performance for off-the-shelf LLMs by selecting critical tokens to reprocess in a nested depth manner. Specifically, at the end of the given transformer layer, DND identifies more critical tokens with a router and feeds them back for an extra round of processing, effectively "reviewing" difficult tokens while avoiding redundant computation for easier ones. The dynamic selection mechanism is tailored for precise control via two novel strategies: a router controlling loss to enhance token selection distinguishability, and a threshold control scheme to ensure selection stability. We demonstrate the effectiveness of DND by directly integrating it into pre-trained dense and MoE models during a post-training phase. On diverse benchmarks, DND boosts the performances of the dense Qwen3-1.7B, Llama3.2-1B, and Gemma3-1B by 1.88%, 2.61%, and 2.50% and the MoE Qwen3-30B-A3B by 0.87%, all with a minimal parameter and computing increase.

Efficient processing of high-resolution images is crucial for real-world vision–language applications. However, existing Large Vision-Language Models (LVLMs) incur substantial computational overhead due to the large number of vision tokens. With the advent of "thinking with images" models, reasoning now extends beyond text to the visual domain. This capability motivates our two-stage "coarse-to-fine" reasoning pipeline: first, a downsampled image is analyzed to identify task-relevant regions; then, only these regions are cropped at full resolution and processed in a subsequent reasoning stage. This approach reduces computational cost while preserving fine-grained visual details where necessary. A major challenge lies in inferring which regions are truly relevant to a given query. Recent related methods often fail in the first stage after input-image downsampling, due to perception-driven reasoning, where clear visual information is required for effective reasoning. To address this issue, we propose ERGO (Efficient Reasoning & Guided Observation) that performs reasoning-driven perception—leveraging multimodal context to determine where to focus. Our model can account for perceptual uncertainty, expanding the cropped region to cover visually ambiguous areas for answering questions. To this end, we develop simple yet effective reward components in a reinforcement learning framework for coarse-to-fine perception. Across multiple datasets, our approach delivers higher accuracy than the original model and competitive methods, with greater efficiency. For instance, ERGO surpasses Qwen2.5-VL-7B on the V* benchmark by 4.7 points while using only 23% of the vision tokens, achieving a 3× inference speedup.

Recently spatial-temporal intelligence of Visual-Language Models (VLMs) has attracted much attention due to its importance for Autonomous Driving, Embodied AI and General Artificial Intelligence. Existing spatial-temporal benchmarks mainly focus on egocentric perspective reasoning with images/video context, or geographic perspective reasoning with graphics context (eg. a map), thus fail to assess VLMs' geographic spatial-temporal intelligence with both images/video and graphics context, which is important for areas like traffic management and emergency response. To address the gaps, we introduce Geo-Temporal Reasoning benchmark (GTR-Bench), a novel challenge for geographic temporal reasoning of moving targets in a large-scale camera network. GTR-Bench is more challenging as it requires multiple perspective switches between maps and videos, joint reasoning across multiple videos with non-overlapping fields of view, and inference over spatial-temporal regions that are unobserved by any video context. Evaluations of more than 10 popular VLMs on GTR-Bench demonstrate that even the best proprietary model, Gemini-2.5-Pro (34.9\%), significantly lags behind human performance (78.61\%) on geo-temporal reasoning. Moreover, our comprehensive analysis on GTR-Bench reveals three primary deficiencies of current models for geo-temporal reasoning. (1) VLMs' reasoning is impaired by an imbalanced utilization of spatial-temporal context. (2) VLMs are weak in temporal forecasting, which leads to worse performance on temporal-emphasized tasks than on spatial-emphasized tasks. (3) VLMs lack the proficiency to comprehend or align the map data with multi-view video inputs. We believe GTR-Bench offers valuable insights and opens up new opportunities for research and applications in spatial-temporal intelligence. Benchmark and code will be released at \href{https://anonymous.4open.science/r/GTR-Bench-5B76}{https://anonymous.4open.science/r/GTR-Bench-5B76}.

Robotic manipulation with Vision-Language-Action models requires efficient inference over long-horizon multi-modal context, where attention to dense visual tokens dominates computational cost. Existing methods optimize inference speed by reducing visual redundancy within VLA models, but they overlook the varying redundancy across robotic manipulation stages. We observe that the visual token redundancy is higher in coarse manipulation phase than in fine-grained operations, and is strongly correlated with the action dynamic. Motivated by this observation, we propose Action-aware Dynamic Pruning (ADP), a multi-modal pruning framework that integrates text-driven token selection with action-aware trajectory gating. ADP introduces a gating mechanism that conditions the pruning signal on recent action trajectories, using past motion windows to adaptively adjust token retention ratios in accordance with dynamics, thereby balancing computational efficiency and perceptual precision across different manipulation stages. Extensive experiments on the LIBERO suites and diverse real-world scenarios demonstrate that our method significantly reduces FLOPs and action inference latency (e.g. 1.35× speed up on OpenVLA-OFT) while maintaining competitive success rates (e.g. 25.8% improvements with OpenVLA) compared to baselines, thereby providing a simple plug-in path to efficient robot policies that advances the efficiency and performance frontier of robotic manipulation.

Generating 3D human poses from multimodal inputs such as text or images requires models to capture both rich semantic and spatial correspondences. While pose-specific multimodal large language models (MLLMs) have shown promise, their supervised fine-tuning (SFT) paradigm struggles to resolve the task's inherent ambiguity. Its reliance on objectives like SMPL parameter regression creates a critical alignment gap, compromising the model's ability to achieve the required semantic and spatial fidelity. To close the gap, we propose Pose-RFT, a framework that shifts the learning paradigm from supervised imitation to reward-driven reinforcement fine-tuning (RFT). We address the core technical challenge of this task: a hybrid action space requiring joint optimization of discrete language and continuous pose outputs. To this end, we introduce HyGRPO, a hybrid reinforcement learning algorithm that enables stable optimization by performing group-wise reward normalization over sampled responses. Pose-RFT incorporates task-specific reward functions to guide optimization towards spatial alignment in image-to-pose generation and semantic consistency in text-to-pose generation. Extensive experiments on multiple pose generation benchmarks demonstrate that Pose-RFT significantly improves performance over existing pose-specific MLLMs, validating the effectiveness of our approach in closing the alignment gap for 3D pose generation.

Recently, we have witnessed great progress in image editing with natural language instructions. Several closed-source models like GPT-Image-1, Seedream, and Google-Nano-Banana have shown highly promising progress. However, the open-source models are still lagging. The main bottleneck is the lack of a reliable reward model to scale up high-quality synthetic training data. To address this critical bottleneck, we built EditReward, trained with our new large-scale human preference dataset, meticulously annotated by trained experts following a rigorous protocol containing over 200K preference pairs. EditReward demonstrates superior alignment with human preferences in instruction-guided image editing tasks. Experiments show that EditReward achieves state-of-the-art human correlation on established benchmarks such as GenAI-Bench, AURORA-Bench, ImagenHub, and our new EditReward-Bench, outperforming a wide range of VLM-as-judge models. Furthermore, we use EditReward to select a high-quality subset from the existing noisy ShareGPT-4o-Image dataset. We train Step1X-Edit on the selected subset, which shows significant improvement over training on the full set. This demonstrates EditReward's ability to serve as a reward model to scale up high-quality training data for image editing. EditReward with its training dataset will be released to help the community build more high-quality image editing training datasets to catch up with the frontier ones.

Transformer-based language models (LMs) notably struggle to reliably capture distant contextual information. This work introduces a novel approach using meta-tokens -- special tokens injected during pre-training -- paired with a dedicated meta-attention mechanism to guide LMs to use these tokens. We pre-train a language model equipped with meta-attention in addition to causal multi-head attention on <100B tokens, achieving strong performance on a suite of synthetic tasks. Our method facilitates length generalization up to 2$\times$ the context window after extension with YaRN. We provide an information-theoretic analysis which reveals that meta-tokens \textit{sharpen} the positional encoding, allowing them to operate as content-based anchors that compress preceding context and “cache” it within the meta-token. We empirically confirm this by visualizing model internals to study the residual stream. Together, our findings demonstrate that meta-tokens and meta-attention provide a simple, data-efficient pre-training method, grounded by new mechanistic insights into their role in enabling length generalization behavior.

View transformers process multi-view observations to predict actions and have shown impressive performance in robotic manipulation. Existing methods typically extract static visual representations in a view-specific manner, leading to inadequate 3D spatial reasoning ability and a lack of dynamic adaptation. Taking inspiration from how the human brain integrates static and dynamic views to address these challenges, we propose Cortical Policy, a novel dual-stream view transformer for robotic manipulation that jointly reasons from static-view and dynamic-view streams. The static-view stream enhances spatial understanding by aligning features of geometrically consistent keypoints extracted from a pretrained 3D foundation model. The dynamic-view stream achieves adaptive adjustment through position-aware pretraining of an egocentric gaze estimation model, computationally replicating the human cortical dorsal pathway. Subsequently, the complementary view representations of both streams are integrated to determine the final actions, enabling the model to handle spatially-complex and dynamically-changing tasks under language conditions. Empirical evaluations on RLBench, the challenging COLOSSEUM benchmark, and real-world tasks demonstrate that Cortical Policy outperforms state-of-the-art baselines substantially, validating the superiority of dual-stream design for visuomotor control. Our cortex-inspired framework offers a fresh perspective for robotic manipulation and holds potential for broader application in vision-based robot control.

As Retrieval-Augmented Generation (RAG) evolves into service-oriented platforms (Rag-as-a-Service) with shared knowledge bases, protecting the copyright of contributed data becomes essential. Existing watermarking methods in RAG focus solely on textual knowledge, leaving image knowledge unprotected. In this work, we propose \textit{AQUA}, the first watermark framework for image knowledge protection in Multimodal RAG systems. \textit{AQUA} embeds semantic signals into synthetic images using two complementary methods: acronym-based triggers and spatial relationship cues. These techniques ensure watermark signals survive indirect watermark propagation from image retriever to textual generator, being efficient, effective and imperceptible. Experiments across diverse models and datasets show that \textit{AQUA} enables robust, stealthy, and reliable copyright tracing, filling a key gap in multimodal RAG protection.

Supervised fine-tuning (SFT) has become the de facto post-training strategy for large vision-language-action (VLA) models, but its reliance on costly human demonstrations limits scalability and generalization. We propose Probe, Learn, Distill (PLD), a plug-and-play framework that improves VLAs through residual reinforcement learning and distribution-aware data collection. In Stage 1 (specialist acquisition), we freeze the VLA backbone and train lightweight residual actors via off-policy RL. These specialists take over in states where the base policy fails, thereby probing failure regions of the generalist. In Stage 2 (data collection), we employ a hybrid rollout scheme that biases residual interventions toward states frequently visited by the base policy, aligning collected trajectories with the generalist’s deployment distribution while capturing recovery behaviors. In Stage 3 (fine-tuning), these curated trajectories are distilled back into the generalist with standard SFT, applicable to both flow-matching and autoregressive heads. We evaluate PLD across diverse settings: it achieves a near-saturated 99% task success rate on the LIBERO benchmark, delivers over 50% performance gains in SimplerEnv, and demonstrates practicality on real-world Franka arm manipulation tasks. We further provide ablations showing that residual policy probing and distribution-aware replay are key to collecting deployment-aligned data that improves VLAs’ capabilities on both seen and unseen tasks. Our results demonstrate that RL-generated, policy-aligned data can surpass teleoperation-only demonstrations, offering a scalable path toward self-improving VLA models.

We introduce enhanced Constitutional Classifiers that deliver production-grade jailbreak robustness with dramatically reduced computational costs and refusal rates compared to previous-generation defenses. We first identify vulnerabilities in existing systems that evaluate model outputs without regard to the conversational context, and address these vulnerabilities using full exchange classifiers. Building on this, we implement a classifier cascade where lightweight classifiers screen all traffic, escalating only suspicious exchanges to more expensive classifiers. Combining this approach with other optimizations, we develop a new production-grade jailbreak defense system that achieves a 5.4× computational cost reduction compared to our baseline exchange classifier, while also achieving a 0.036% refusal rate on production traffic. Through extensive red-teaming comprising over 560K queries, we demonstrate protection against universal jailbreaks---no attack on this system successfully elicited responses to all eight target queries comparable in detail to an undefended model. Finally, we explore efficient classification techniques by training linear activation probes. We show using logit smoothing and a weighted loss function is crucial for performance, and further that probes can be combined with external classifiers to provide particularly strong performance. Our work establishes Constitutional Classifiers as practical safeguards for large language models.

Large vision–language models (VLMs) excel at multimodal understanding but fall short when extended to embodied tasks, where instructions must be transformed into low-level motor actions. We introduce SP-VLA, a dual-system **V**ision–**L**anguage–**A**ction framework that leverages **S**patial **P**riors as a bridge between linguistic instructions and embodiment-specific control. introduce SP-VLA aligns action learning with spatial priors through two stages: (i) spatial grounding pre-training, which equips the VLM with transferable priors via scalable point, box, and trajectory prediction from both web-scale and robot-specific data, and (ii) spatially guided action post-training, which encourages the model to produce richer spatial priors to guide action generation via spatial prompting. This design preserves spatial grounding during policy learning and promotes consistent optimization across spatial and action objectives. Empirically, introduce SP-VLA achieves substantial improvements over vanilla VLA, with performance increasing from $66.1{\rightarrow}84.6$ on Google Robot and from $54.7{\rightarrow}73.2$ on WidowX, establishing new state-of-the-art results on SimplerEnv. It also demonstrates stronger generalization to unseen objects and paraphrased instructions, as well as robustness to long-horizon perturbations in real-world settings. These results highlight scalable spatially guided training as a promising direction for robust, generalizable robot learning. We will release code, data, and model checkpoints to support future research. See more visualization results at the anonymous page: https://sp-vla-anonymous.vercel.app

In complex embodied long-horizon manipulation tasks, effective task decomposition and execution require synergistic integration of textual logical reasoning and visual-spatial imagination to ensure efficient and accurate operation. Current methods fail to adopt a unified generation framework for multimodal planning, leading to inconsistencies in multimodal planning. To address this challenge, we present EVLP (Embodied Vision-Language Planner), an innovative multimodal unified generation framework that jointly models linguistic reasoning and visual generation. Our approach achieves multimodal planning for long-horizon tasks through a novel training pipeline incorporating dynamic pretraining and reinforced alignment. Our core innovations consist of three key components: 1. Unified Multimodal Generation Framework: For understanding, we integrate semantic information with spatial features to provide comprehensive visual perception. For generation, we directly learn the joint distribution of discrete images for one-step visual synthesis, enabling coordinated language-visual modeling through learnable cross-modal attention mechanisms. 2. Dynamic Perception Pretraining: We propose a bidirectional dynamic alignment strategy employing inverse dynamics tasks and forward dynamics tasks, effectively strengthening multimodal correlations within a unified feature space. 3. Reinforced Supervised Fine-Tuning: While conducting instruction-based fine-tuning in the unified generation space, we construct a reinforce loss to align the spatial logic between textual actions and generated images, enabling the model to acquire spatio-aware multimodal planning capabilities.Comprehensive evaluations on multiple complex tasks demonstrate that EVLP significantly outperforms competitive baselines in both instruction execution accuracy and task success rate, benefiting from its unified multimodal architecture and well-designed training pipeline. Extensive ablation studies further validate the rationality of our framework design.