ICLR 2026 论文列表

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.

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.

Text-to-Motion generation has become a fundamental task in human-machine interaction, enabling the synthesis of realistic human motions from natural language descriptions. Although recent advances in large language models and reinforcement learning have contributed to high-quality motion generation, two major challenges remain. Existing approaches often fail to capture the temporal and causal complexities inherent in natural language, leading to oversimplified or incoherent motions. Additionally, RL-based methods are frequently overly complex, hindering their scalability and adaptability across various motion generation tasks. To address these challenges, we propose **Motion-R1**, a novel framework that combines decomposed Chain-of-Thought reasoning with reinforcement learning to enhance both the quality and interpretability of generated motions. Specifically, we introduce the **Decomposed CoT Data Engine**, which leverages an automated pipeline to synthesize high-quality reasoning data, allowing the model to better capture the temporal dependencies and causal relationships of human motion. We also propose **RL Binding**, a reinforcement learning strategy that incorporates multi-modal text-motion alignment into the RL reward function, guiding the model to produce motions that are both semantically accurate and motionally realistic. Extensive experiments across benchmark datasets demonstrate that Motion-R1 achieves state-of-the-art performance, with a 3.5\% improvement in MM-Dist on HumanML3D and improvements in R-Precision and FID on KIT-ML and BABEL, surpassing existing methods across key metrics and highlighting its superior capability in handling complex motion generation tasks.

Large Language Models (LLMs) have demonstrated impressive reasoning capabilities in complex problem-solving tasks, sparking growing interest in their application to preference reasoning in recommendation systems. Existing methods typically rely on fine-tuning with explicit chain-of-thought (CoT) data. However, these methods face significant practical limitations due to (1) the difficulty of obtaining high-quality CoT data in recommendation and (2) the high inference latency caused by generating CoT reasoning. In this work, we explore an alternative approach that shifts from explicit CoT reasoning to compact, information-dense latent reasoning. This approach eliminates the need for explicit CoT generation and improves inference efficiency, as a small set of latent tokens can effectively capture the entire reasoning process. Building on this idea, we propose *Reinforced Latent Reasoning for Recommendation* (LatentR$^3$), a novel end-to-end training framework that leverages reinforcement learning (RL) to optimize latent reasoning without relying on any CoT data.LatentR$^3$ adopts a two-stage training strategy: first, supervised fine-tuning to initialize the latent reasoning module, followed by pure RL training to encourage exploration through a rule-based reward design. Our RL implementation is based on a modified GRPO algorithm, which reduces computational overhead during training and introduces continuous reward signals for more efficient learning. Extensive experiments demonstrate that LatentR$^3$ enables effective latent reasoning without any direct supervision of the reasoning process, significantly improving performance when integrated with different LLM-based recommendation methods. Our codes are available at https://anonymous.4open.science/r/R3-A278/.

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.

Recent advances in offline reinforcement learning (RL) have led to the development of high-performing algorithms that achieve impressive results across standard benchmarks. However, many of these methods depend on increasingly complex planning architectures, which hinder their deployment in real-world settings due to high inference costs. To overcome this limitation, recent research has explored data augmentation techniques that offload computation from online decision-making to offline data preparation. Among these, diffusion-based generative models have shown potential in synthesizing diverse trajectories but incur significant overhead in training and data generation. In this work, we propose Trajectory Generation with Conservative Value Guidance (TGCVG), a novel trajectory-level data augmentation framework that integrates a high-performing offline policy with a learned dynamics model. To ensure that the synthesized trajectories are both high-quality and close to the original dataset distribution, we introduce a value-guided regularization during the training of the offline policy. This regularization encourages conservative action selection, effectively mitigating distributional shift during trajectory synthesis. Empirical results on standard benchmarks demonstrate that TGCVG not only improves the performance of state-of-the-art offline RL algorithms but also significantly reduces training and trajectory synthesis time. These findings highlight the effectiveness of value-aware data generation in improving both efficiency and policy performance.

Scaling imitation learning to high-DoF whole-body robots is fundamentally limited by the \textbf{curse of dimensionality} and the prohibitive cost of collecting expert demonstrations. We argue that the core bottleneck is paradigmatic: real-world supervision for whole-body control is inherently imperfect, yet most methods assume expert data. To overcome this, we propose \textbf{HVD} (Hierarchical Value-Decomposed Offline Reinforcement Learning), an offline reinforcement learning framework that learns effective policies directly from suboptimal, reward-labeled trajectories. HVD structures the value function along the robot’s kinematic hierarchy and over temporal chunks, enabling precise credit assignment in long-horizon, high-dimensional tasks. Built on a Transformer-based architecture, HVD supports \textit{multi-modal} and \textit{multi-task} learning, allowing flexible integration of diverse sensory inputs for policy learning. To enable realistic evaluation and training, we further introduce \textbf{WB-50}, a 50-hour dataset of teleoperated and policy rollout trajectories annotated with rewards and preserving natural imperfections — including partial successes, corrections, and failures. Experiments show HVD significantly outperforms existing baselines in success rate across complex whole-body tasks. Our results suggest that effective policy learning for high-DoF systems can emerge not from perfect demonstrations, but from structured learning over realistic, imperfect data.

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.

The fast-growing demands in using Large Language Models (LLMs) to tackle complex multi-step data science tasks create a emergent need for accurate benchmarking. There are two major gaps in existing benchmarks: (i) the lack of standardized, process-aware evaluation that captures instruction adherence and process fidelity, and (ii) the scarcity of accurately labeled training data. To bridge these gaps, we introduce DARE-bench, a benchmark designed for machine learning modeling and data science instruction following. Unlike many existing benchmarks that rely on human- or model-based judges, all tasks in DARE-bench have verifiable ground truth, ensuring objective and reproducible evaluation. To cover a broad range of tasks and support agentic tools, DARE-bench consists of 6,300 Kaggle-derived tasks and provides both large-scale training data and evaluation sets. Extensive evaluations show that even highly capable models such as gpt-o4-mini struggle to achieve good performance, especially in machine learning modeling tasks. Using DARE-bench training tasks for fine-tuning can substantially improve model performance. For example, supervised fine-tuning boosts Qwen3-32B’s accuracy by 1.83× and reinforcement learning boosts Qwen3-4B’s accuracy by more than 8×. These significant improvements verify the importance of DARE-bench both as an accurate evaluation benchmark and critical training data.

Reinforcement learning with verifiable rewards (RLVR) has become a trending paradigm for training reasoning large language models (LLMs). However, due to the autoregressive decoding nature of LLMs, the rollout process becomes the efficiency bottleneck of RL training, consisting of up to 70\% of the total training time. In this work, we propose Quantized Reinforcement Learning (QuRL) that uses a quantized actor for accelerating the rollout. We address two challenges in QuRL. First, we propose Adaptive Clipping Range (ACR) that dynamically adjusts the clipping ratio based on the policy ratio between the full-precision actor and the quantized actor, which is essential for mitigating long-term training collapse. Second, we identify the weight update problem, where weight changes between RL steps are extremely small, making it difficult for the quantization operation to capture them effectively. We mitigate this problem through the invariant scaling technique that reduces quantization noise and increases weight update. We evaluate our method with INT8 and FP8 quantization experiments on DeepScaleR and DAPO, and achieve 20% to 80% faster rollout during training.

Fine-tuning Large Language Models (LLMs) with reinforcement learning (RL) has become a key technique for enhancing performance on a wide range of tasks, from user alignment to complex reasoning. However, this approach is often hindered by the difficulty of fine-grained credit assignment, as it typically relies on sparse rewards given only at the end of a completely generated sequence. Conventional solutions often require training an auxiliary value network known as critic, which introduces significant computational overhead and training instability. We present Group-Normalized Implicit Value Optimization (GN-IVO), a novel, critic-free algorithm that directly addresses this challenge. GN-IVO learns step-level values implicitly from the policy through a group-normalized distributional matching objective. This approach elegantly circumvents the need for an explicit critic and avoids the computation of the intractable partition function by normalizing values across a group of sampled model responses. Theoretically, we prove that our objective recovers the true value function up to a constant, guaranteeing that the optimal policy is preserved. We demonstrate the practical effectiveness of GN-IVO on a diverse set of text generation and reasoning tasks, showing that it consistently outperforms strong RL baselines for LLMs.

Search-augmented large language models (LLMs) trained with reinforcement learning (RL) have achieved strong results on open-domain question answering (QA), but training still remains a significant challenge. The optimization is often unstable due to sparse rewards and difficult credit assignments across reasoning and tool calls. To address this, we introduce Turn-Level Information Potential Reward Shaping (TIPS), a simple framework that assigns dense, turn-level rewards to each reasoning + tool-call segment based on the increased likelihood of the correct answer under a teacher model. By leveraging the potential-based reward shaping, TIPS offers fine-grained and policy-invariant guidance that overcomes the limitations of outcome-only optimization. Evaluated on seven QA benchmarks, TIPS consistently outperforms GRPO/PPO baselines and substantially improves training stability. For instance, with a Qwen-2.5 7B Instruct model, TIPS improves the average Exact Match score by 11.8% and F1 by 13.6% relative to PPO. Our results demonstrate that turn-level information-potential reward shaping provides an effective and general solution to sparse-reward credit assignment for multi-turn LLM reasoning.

In Reward Learning (ReL), we are given feedback on an unknown target reward, and the goal is to use this information to recover it in order to carry out some downstream application, e.g., planning. When the feedback is not informative enough, the target reward is only partially identifiable, i.e., there exists a set of rewards, called the feasible set, that are equally plausible candidates for the target reward. In these cases, the ReL algorithm might recover a reward function different from the target reward, possibly leading to a failure in the application. In this paper, we introduce a general ReL framework that permits to quantify the drop in "performance" suffered in the considered application because of identifiability issues. Building on this, we propose a robust approach to address the identifiability problem in a principled way, by maximizing the "performance" with respect to the worst-case reward in the feasible set. We then develop Rob-ReL, a ReL algorithm that applies this robust approach to the subset of ReL problems aimed at assessing a preference between two policies, and we provide theoretical guarantees on sample and iteration complexity for Rob-ReL. We conclude with some numerical simulations to illustrate the setting and empirically characterize Rob-ReL.

Vision-language reinforcement learning (RL) has primarily focused on narrow domains (e.g. geometry or chart reasoning). This leaves broader training scenarios and resources underexplored, limiting the exploration and learning of Vision Language Models (VLMs) through RL. We find video games inherently provide rich visual elements and mechanics that are easy to verify. To fully use the multimodal and verifiable reward in video games, we propose Game-RL, constructing diverse game tasks for RL training to boost VLMs general reasoning ability. To obtain training data, we propose Code2Logic, a novel approach that adapts game code to synthesize game reasoning task data, thus obtaining the GameQA dataset of 30 games and 158 tasks with controllable difficulty gradation. Unexpectedly, training solely on GameQA would help VLMs obtain better out of domain generalization, demonstrating the value of Game-RL for enhancing VLMs general reasoning. Furthermore, this suggets that RL can lead to generalizable improvements in VLMs' reasoning abilities, and notably, video games may serve as valuable scenarios and resources to bring this generalization.

Large Reasoning Models (LRMs) like o3 and DeepSeek-R1 have achieved remarkable progress in natural language reasoning with long chain-of-thought. However, they remain computationally inefficient and struggle with accuracy when solving problems requiring complex mathematical operations. In this work, we present AgentMath, an agent framework that seamlessly integrates language models' reasoning capabilities with code interpreters' computational precision to efficiently tackle complex mathematical problems. Our approach introduces three key innovations: (1) An automated method that converts natural language chain-of-thought into structured tool-augmented trajectories, generating high-quality supervised fine-tuning (SFT) data to alleviate data scarcity; (2) A novel agentic reinforcement learning (RL) paradigm that dynamically interleaves natural language generation with real-time code execution. This enables models to autonomously learn optimal tool-use strategies through multi-round interactive feedback, while fostering emergent capabilities in code refinement and error correction; (3) An efficient training system incorporating innovative techniques, including request-level asynchronous rollout scheduling, agentic partial rollout, and prefix-aware weighted load balancing, achieving 4-5× speedup and making efficient RL training feasible on ultra-long sequences with scenarios with massive tool calls. Extensive evaluations show that AgentMath achieves state-of-the-art performance on challenging mathematical competition benchmarks including AIME24, AIME25, and HMMT25, substantially outperforming frontier open‑source models of comparable size. Specifically, AgentMath-30B-A3B attains 90.6\%, 86.4\%, and 73.8\% accuracy respectively, surpassing OpenAI-o3-mini and Claude-Opus-4.0-Thinking while remaining competitive with OpenAI-o3, Gemini-2.5-Pro, and DeepSeek-R1-671B-0528. These results validate the effectiveness of our approach and pave the way for building more sophisticated and scalable mathematical reasoning agents.

Speculative decoding accelerates large language model (LLM) inference by letting a lightweight draft model propose multiple tokens that the target model verifies in parallel. Yet existing training objectives optimize only a single greedy draft path, while decoding follows a tree policy that re-ranks and verifies multiple branches. This draft policy misalignment limits achievable speedups. We introduce Group Tree Optimization (GTO), which aligns training with the decoding-time tree policy through two components: (i) Draft Tree Reward, a sampling-free objective equal to the expected acceptance length of the draft tree under the target model, directly measuring decoding performance; (ii) Group-based Draft Policy Training, a stable optimization scheme that contrasts trees from the current and a frozen reference draft model, forming debiased group-standardized advantages and applying a PPO-style surrogate along the longest accepted sequence for robust updates. We further prove that increasing our Draft Tree Reward provably improves acceptance length and speedup. Across dialogue (MT-Bench), code (HumanEval), and math (GSM8K), and multiple LLMs (e.g., LLaMA-3.1-8B, LLaMA-3.3-70B, Vicuna-1.3-13B, DeepSeek-R1-Distill-LLaMA-8B), GTO increases acceptance length by 7.4\% and yields an additional 7.7\% speedup over prior state-of-the-art EAGLE-3. By bridging draft policy misalignment, GTO offers a practical, general solution for efficient LLM inference. Code and draft models are available at https://anonymous.4open.science/r/GTO-ICLR-348F/.

Recent advancements in reinforcement learning with verifiable rewards have pushed the boundaries of the visual reasoning capabilities in large vision-language models (LVLMs). However, training LVLMs with reinforcement fine-tuning (RFT) is computationally expensive, posing a significant challenge to scaling model size. In this work, we propose ProxyThinker, an inference-time technique that enables large models to inherit the visual reasoning capabilities from small, slow-thinking visual reasoners without any training. By subtracting the output distributions of base models from those of RFT reasoners, ProxyThinker modifies the decoding dynamics and successfully elicits the slow-thinking reasoning demonstrated by the emerged sophisticated behaviors, such as self-verification and self-correction. ProxyThinker consistently boosts performance on challenging visual benchmarks on spatial, mathematical, and multidisciplinary reasoning, enabling untuned base models to compete with the performance of their full-scale RFT counterparts. Furthermore, our implementation efficiently coordinates multiple language models with parallelism techniques and achieves faster inference compared to previous decoding-time methods, paving the way for the practical deployment of ProxyThinker. Code is available at https://github.com/MrZilinXiao/ProxyThinker.

GUI grounding, which maps natural-language instructions to actionable UI elements, is a core capability of GUI agents. Prior work largely treats instructions as a static proxy for user intent, overlooking the impact of instruction diversity on grounding performance. Through a careful investigation of existing grounding datasets, we find a 23.3\% flaw rate in their instructions and show that inference-time exploitation of instruction diversity yields up to a 76\% relative performance improvement. In this paper, we introduce the "Instruction as Reasoning" paradigm, treating instructions as dynamic analytical pathways that offer distinct perspective and enabling the model to select the most effective pathway during reasoning. To achieve this, we propose a two-stage training framework: supervised fine-tuning (SFT) on synthesized, diverse instructions to instill multi-perspective reasoning, followed by reinforcement learning (RL) to optimize pathway selection and composition. Our resulting models, UI-Ins-7B and UI-Ins-32B, achieve state-of-the-art results on five challenging benchmarks and exhibit emergent reasoning, selectively composing and synthesizing novel instruction pathways at inference. In particular, UI-Ins-32B attains the best grounding accuracy: 87.3\% on UI-I2E-Bench and 84.9\% on MMBench-GUI L2, besides, UI-Ins-7B yields superior agent performance, achieving a 66.1\% success rate on the AndroidWorld. All code, data, and models will be publicly released.

A fundamental challenge in offline reinforcement learning is distributional shift. Scarce data or datasets dominated by out-of-distribution (OOD) areas exacerbate this issue. Our theoretical analysis and experiments show that the standard squared error objective induces a harmful TD cross covariance. This effect amplifies in OOD areas, biasing optimization and degrading policy learning. To counteract this mechanism, we develop two complementary strategies: partitioned buffer sampling that restricts updates to localized replay partitions, attenuates irregular covariance effects, and aligns update directions, yielding a scheme that is easy to integrate with existing implementations, namely Clustered Cross-Covariance Control for TD ($C^4$). We also introduce an explicit gradient-based corrective penalty that cancels the covariance induced bias within each update. We prove that buffer partitioning preserves the lower bound property of the maximization objective, and that these constraints mitigate excessive conservatism in extreme OOD areas without altering the core behavior of policy constrained offline reinforcement learning. Empirically, our method showcases higher stability and up to 30% improvement in returns over prior methods, especially with small datasets and splits that emphasize OOD areas.