NeurIPS 2025 论文列表

Clinical decision‑making routinely demands reasoning over heterogeneous data, yet existing multimodal language models (MLLMs) remain largely vision‑centric and fail to generalize across clinical specialties. To bridge this gap, we introduce QoQ-Med-7B/32B, the first open generalist clinical foundation model that jointly reasons across medical images, time‑series signals, and text reports. QoQ-Med is trained with Domain‑aware Relative Policy Optimization (DRPO), a novel reinforcement‑learning objective that hierarchically scales normalized rewards according to domain rarity and modality difficulty, mitigating performance imbalance caused by skewed clinical data distributions. Trained on 2.61 million instruction tuning pairs spanning 9 clinical domains, we show that DRPO training boosts diagnostic performance by 43% in macro‑F1 on average across all visual domains as compared to other critic-free training methods like GRPO. Furthermore, with QoQ-Med trained on intensive segmentation data, it is able to highlight salient regions related to the diagnosis, with an IoU 10x higher than open models while reaching the performance of OpenAI o4-mini. To foster reproducibility and downstream research, we release (i) the full model weights, (ii) the modular training pipeline, and (iii) all intermediate reasoning traces.

We study offline policy optimization for infinite-horizon average-reward Markov decision processes (MDPs) with large or infinite state spaces. Specifically, we propose a pessimistic actor-critic algorithm that uses a computationally efficient linear function class for value function estimation. At the core of our method is a critic that computes a pessimistic estimate of the average reward under the current policy, as well as the corresponding policy gradient, by solving a fixed-point Bellman equation, rather than solving a successive sequence of regression problems as in finite horizon settings. This procedure reduces to solving a second-order cone program, which is computationally tractable. Our theoretical analysis is based on a weak partial data coverage assumption, which requires only that the offline data aligns well with the expected feature vector of a comparator policy. Under this condition, we show that our algorithm achieves the optimal sample complexity of O(\varepsilon^{-2}) for learning a near-optimal policy, up to model misspecification errors.

This paper studies the problem of safe meta-reinforcement learning (safe meta-RL), where an agent efficiently adapts to unseen tasks while satisfying safety constraints at all times during adaptation. We propose a framework consisting of two complementary modules: safe policy adaptation and safe meta-policy training. The first module introduces a novel one-step safe policy adaptation method that admits a closed-form solution, ensuring monotonic improvement, constraint satisfaction at every step, and high computational efficiency. The second module develops a Hessian-free meta-training algorithm that incorporates safety constraints on the meta-policy and leverages the analytical form of the adapted policy to enable scalable optimization. Together, these modules yield three key advantages over existing safe meta-RL methods: (i) superior optimality, (ii) anytime safety guarantee, and (iii) high computational efficiency. Beyond existing safe meta-RL analyses, we prove the anytime safety guarantee of policy adaptation and provide a lower bound of the expected total reward of the adapted policies compared with the optimal policies, which shows that the adapted policies are nearly optimal. Empirically, our algorithm achieves superior optimality, strict safety compliance, and substantial computational gains—up to 70\% faster training and 50\% faster testing—across diverse locomotion and navigation benchmarks.

Optimal decision-making under partial observability requires agents to balance reducing uncertainty (exploration) against pursuing immediate objectives (exploitation). In this paper, we introduce a novel policy optimization framework for continuous partially observable Markov decision processes (POMDPs) that explicitly addresses this challenge. Our method casts policy learning as probabilistic inference in a non-Markovian Feynman--Kac model that inherently captures the value of information gathering by anticipating future observations, without requiring suboptimal approximations or handcrafted heuristics. To optimize policies under this model, we develop a nested sequential Monte Carlo (SMC) algorithm that efficiently estimates a history-dependent policy gradient under samples from the optimal trajectory distribution induced by the POMDP. We demonstrate the effectiveness of our algorithm across standard continuous POMDP benchmarks, where existing methods struggle to act under uncertainty.

Garment manipulation is a critical challenge due to the diversity in garment categories, geometries, and deformations. Despite this, humans can effortlessly handle garments, thanks to the dexterity of our hands. However, existing research in the field has struggled to replicate this level of dexterity, primarily hindered by the lack of realistic simulations of dexterous garment manipulation. Therefore, we propose DexGarmentLab, the first environment specifically designed for dexterous (especially bimanual) garment manipulation, which features large-scale high-quality 3D assets for 15 task scenarios, and refines simulation techniques tailored for garment modeling to reduce the sim-to-real gap. Previous data collection typically relies on teleoperation or training expert reinforcement learning (RL) policies, which are labor-intensive and inefficient. In this paper, we leverage garment structural correspondence to automatically generate a dataset with diverse trajectories using only a single expert demonstration, significantly reducing manual intervention. However, even extensive demonstrations cannot cover the infinite states of garments, which necessitates the exploration of new algorithms. To improve generalization across diverse garment shapes and deformations, we propose a Hierarchical gArment-manipuLation pOlicy (HALO). It first identifies transferable affordance points to accurately locate the manipulation area, then generates generalizable trajectories to complete the task. Through extensive experiments and detailed analysis of our method and baseline, we demonstrate that HALO consistently outperforms existing methods, successfully generalizing to previously unseen instances even with significant variations in shape and deformation where others fail. Our project page is available at: https://wayrise.github.io/DexGarmentLab/.

Recent advances in diffusion language models (DLMs) have presented a promising alternative to traditional autoregressive large language models (LLMs). However, DLMs still lag behind LLMs in reasoning performance, especially as the number of denoising steps decreases. Our analysis reveals that this shortcoming arises primarily from the independent generation of masked tokens across denoising steps, which fails to capture the token correlation. In this paper, we define two types of token correlation: intra-sequence correlation and inter-sequence correlation, and demonstrate that enhancing these correlations improves reasoning performance. To this end, we propose a Multi-Reward Optimization (MRO) approach, which encourages DLMs to consider the token correlation during the denoising process. More specifically, our MRO approach leverages test-time scaling, reject sampling, and reinforcement learning to directly optimize the token correlation with multiple elaborate rewards. Additionally, we introduce group step and importance sampling strategies to mitigate reward variance and enhance sampling efficiency. Through extensive experiments, we demonstrate that MRO not only improves reasoning performance but also achieves significant sampling speedups while maintaining high performance on reasoning benchmarks.

Visual navigation is a fundamental problem in embodied AI, yet practical deployments demand long-horizon planning capabilities to address multi-objective tasks. A major bottleneck is data scarcity: policies learned from limited data often overfit and fail to generalize OOD. Existing neural network-based agents typically increase architectural complexity that paradoxically become counterproductive in the small-sample regime. This paper introduce NeuRO, a integrated learning-to-optimize framework that tightly couples perception networks with downstream task-level robust optimization. Specifically, NeuRO addresses core difficulties in this integration: (i) it transforms noisy visual predictions under data scarcity into convex uncertainty sets using Partially Input Convex Neural Networks (PICNNs) with conformal calibration, which directly parameterize the optimization constraints; and (ii) it reformulates planning under partial observability as a robust optimization problem, enabling uncertainty-aware policies that transfer across environments. Extensive experiments on both unordered and sequential multi-object navigation tasks demonstrate that NeuRO establishes SoTA performance, particularly in generalization to unseen environments. Our work thus presents a significant advancement for developing robust, generalizable autonomous agents.

Robotic manipulation systems benefit from complementary sensing modalities, where each provides unique environmental information. Point clouds capture detailed geometric structure, while RGB images provide rich semantic context. Current point cloud methods struggle to capture fine-grained detail, especially for complex tasks, which RGB methods lack geometric awareness, which hinders their precision and generalization. We introduce PointMapPolicy, a novel approach that conditions diffusion policies on structured grids of points without downsampling. The resulting data type makes it easier to extract shape and spatial relationships from observations, and can be transformed between reference frames. Yet due to their structure in a regular grid, we enable the use of established computer vision techniques directly to 3D data. Using xLSTM as a backbone, our model efficiently fuses the point maps with RGB data for enhanced multi-modal perception. Through extensive experiments on the RoboCasa and CALVIN benchmarks and real robot evaluations, we demonstrate that our method achieves state-of-the-art performance across diverse manipulation tasks. The overview and demos are available on our project page: https://point-map.github.io/Point-Map/

Recent advances have demonstrated the effectiveness of Reinforcement Learning (RL) in improving the reasoning capabilities of Large Language Models (LLMs). However, existing works inevitably rely on high-quality instructions and verifiable rewards for effective training, both of which are often difficult to obtain in specialized domains. In this paper, we propose Self-play Reinforcement Learning (SeRL) to bootstrap LLM training with limited initial data. Specifically, SeRL comprises two complementary modules: self-instruction and self-rewarding. The former module generates additional instructions based on the available data at each training step, employing comprehensive online filtering strategies to ensure instruction quality, diversity, and difficulty. The latter module introduces a simple yet effective majority-voting mechanism to estimate response rewards for additional instructions, eliminating the need for external annotations. Finally, SeRL performs conventional RL based on the generated data, facilitating iterative self-play learning. Extensive experiments on various reasoning benchmarks and across different LLM backbones demonstrate that the proposed SeRL yields results superior to its counterparts and achieves performance on par with those obtained by high-quality data with verifiable rewards. Our code is available at https://github.com/wantbook-book/SeRL.

Achieving general agile whole-body control on humanoid robots remains a major challenge due to diverse motion demands and data conflicts. While existing frameworks excel in training single motion-specific policies, they struggle to generalize across highly varied behaviors due to conflicting control requirements and mismatched data distributions. In this work, we propose BumbleBee (BB), an expert-generalist learning framework that combines motion clustering and sim-to-real adaptation to overcome these challenges. BB first leverages an autoencoder-based clustering method to group behaviorally similar motions using motion features and motion descriptions. Expert policies are then trained within each cluster and refined with real-world data through iterative delta action modeling to bridge the sim-to-real gap. Finally, these experts are distilled into a unified generalist controller that preserves agility and robustness across all motion types. Experiments on two simulations and a real humanoid robot demonstrate that BB achieves state-of-the-art general whole-body control, setting a new benchmark for agile, robust, and generalizable humanoid performance in the real world.

Swimming in nature achieves remarkable performance through diverse morphological adaptations and intricate solid-fluid interaction, yet exploring this capability in artificial soft swimmers remains challenging due to the high-dimensional control complexity and the computational cost of resolving hydrodynamic details. Traditional approaches often rely on morphology-dependent heuristics and simplified fluid models, which constrain exploration and preclude advanced strategies like vortex exploitation. To address this, we propose an automated framework that combines a unified, reduced-mode control space with a high-fidelity GPU-accelerated simulator. Our control space naturally captures deformation patterns for diverse morphologies, minimizing manual design, while our simulator efficiently resolves the crucial fluid-structure interactions required for learning. We evaluate our method on a wide range of morphologies, from bio-inspired to unconventional. From this general framework, high-performance swimming patterns emerge that qualitatively reproduce canonical gaits observed in nature without requiring domain-specific priors, where state-of-the-art baselines often fail, particularly on complex topologies like a torus. Our work lays a foundation for future opportunities in automated co-design of soft robots in complex hydrodynamic environments. The code is available at https://github.com/changyu-hu/FreeFlow.

We prove that the classic REINFORCE stochastic policy gradient (SPG) method converges to globally optimal policies in finite-horizon Markov Decision Processes (MDPs) with $\textit{any}$ constant learning rate. To avoid the need for small or decaying learning rates, we introduce two key innovations in the stochastic bandit setting, which we then extend to MDPs. $\textbf{First}$, we identify a new exploration property of SPG: the online SPG method samples every action infinitely often (i.o.), improving on previous results that only guaranteed at least two actions would be sampled i.o. This means SPG inherently achieves asymptotic exploration without modification. $\textbf{Second}$, we eliminate the assumption of unique mean reward values, a condition that previous convergence analyses in the bandit setting relied on, but that does not translate to MDPs. Our results deepen the theoretical understanding of SPG in both bandit problems and MDPs, with a focus on how it handles the exploration-exploitation trade-off when standard optimization and stochastic approximation methods cannot be applied, as is the case with large constant learning rates.

Reinforcement learning (RL) has become an indispensable post-training step for unlocking the full potential of Large Language Models (LLMs). Its core motivation is to incentivize the model’s inference trajectory via a reward model, effectively balancing the exploration–exploitation trade-off in scenarios where collecting exhaustive input–output ground-truth pairs is infeasible. This motivation naturally extends to visual generation, where perfect alignment between an image and a textual prompt is inherently ambiguous and often unattainable. However, existing visual generative models are not yet ready for RL due to the following two fundamental drawbacks that undermine the foundations of RL: 1) For diffusion-based models, the actual generation trajectories of sampled images cannot be reliably rewarded, as diffusion inversion is notoriously difficult. 2) For autoregressive (AR) models, we show that the widely used spatial visual tokens do not satisfy the Bellman equation and thus violate the policy improvement theorem of RL. To this end, we propose to use Selftok (Self-consistency Tokenizer), which represents each image as a sequential 1D stream of discrete, autoregressive tokens. Together with language, we train a pure AR vision-language model (VLM) for visual generation. Impressively, without using any text-image training pairs, a simple policy gradient algorithm applied to Selftok tokens significantly boosts visual generation performance, surpassing existing models by a large margin. Implementation details are provided in the Appendix.

This paper investigates the role of explicit thinking process in rule-based reinforcement fine-tuning (RFT) for multi-modal large language models (MLLMs). We first extend \textit{Thinking-RFT} to image classification task, using verifiable rewards for fine-tuning~(FT). Experiments show {Thinking-RFT} significantly outperforms supervised FT and yields a cross-dataset generalization effect. We then rethink and question whether explicit thinking in RFT is always necessary and beneficial. Challenging the convention that explicit thinking is crucial for the success of RFT, we introduce \textit{No-Thinking-RFT}, exploring RFT without thinking by introducing a simple equality accuracy reward. We evaluate No-Thinking-RFT on six diverse tasks across different model sizes and types. Experiment results reveal four key findings: \textbf{(1).} Visual perception tasks do not require thinking during RFT, as No-Thinking-RFT consistently outperforms or matches Thinking-RFT across model sizes and types. \textbf{(2).} Models with limited capabilities struggle to generate high-quality CoT for RFT, making Thinking-RFT less effective than No-Thinking-RFT. \textbf{(3).} There are inconsistencies between the answers in the thinking tags and answer tags for some responses of Thinking-RFT, which show lower average accuracy than the overall accuracy. \textbf{(4).} The performance gain of No-Thinking-RFT mainly stems from improved learning during no thinking FT and the avoidance of inference overthinking, as evidenced by the partial gains from appending empty thinking tags at inference time of Thinking-RFT. We hypothesize that explicit thinking before verifiable answers may hinder reward convergence and reduce performance in certain scenarios. To test this, we propose \textit{Think-After-Answer}, which places thinking after the answer to mitigate this effect for experimental verification. Lastly, we conduct a pilot study to explore whether MLLMs can learn when to think during RFT, introducing an \textit{Adaptive-Thinking} method. Experiments show that model converges to either thinking or not depending on model capability, achieving comparable or better performance than both Thinking and No-Thinking-RFT. Our findings suggest MLLMs can adaptively decide to think or not based on their capabilities and task complexity, offering insights into the thinking process in RFT.

Imitation learning seeks policies reflecting the values of demonstrated behaviors. Prevalent approaches learn to match or exceed the demonstrator's performance in expectation without knowing the demonstrator’s reward function. Unfortunately, this does not induce pluralistic imitators that learn to support qualitatively distinct demonstrations. We reformulate imitation learning using stochastic dominance over the demonstrations' reward distribution across a range of reward functions as our foundational aim. Our approach matches imitator policy samples (or support) with demonstrations using optimal transport theory to define an imitation learning objective over trajectory pairs. We demonstrate the benefits of pluralistic stochastic dominance (PSD) for imitation in both theory and practice.

In this paper, we explore semantic clustering properties of deep reinforcement learning (DRL) to improve its interpretability and deepen our understanding of its internal semantic organization. In this context, semantic clustering refers to the ability of neural networks to cluster inputs based on their semantic similarity in the feature space. We propose a DRL architecture that incorporates a novel semantic clustering module that combines feature dimensionality reduction with online clustering. This module integrates seamlessly into the DRL training pipeline, addressing the instability of t-SNE and eliminating the need for extensive manual annotation inherent to prior semantic analysis methods. We experimentally validate the effectiveness of the proposed module and demonstrate its ability to reveal semantic clustering properties within DRL. Furthermore, we introduce new analytical methods based on these properties to provide insights into the hierarchical structure of policies and semantic organization within the feature space. Our code is available at https://github.com/ualiangzhang/semantic_rl.

Recent advances in reasoning-centric language models have highlighted reinforcement learning (RL) as a promising method for aligning models with verifiable rewards. However, it remains contentious whether RL truly expands a model’s reasoning capabilities or merely amplifies high-reward outputs already latent in the base model’s distribution, and whether continually scaling up RL compute reliably leads to improved reasoning performance. In this work, we challenge prevailing assumptions by demonstrating that prolonged RL (ProRL) training can uncover novel reasoning strategies that are inaccessible to base models, even under extensive sampling. We introduce ProRL, a novel training methodology that incorporates KL divergence control, reference policy resetting, and a diverse suite of tasks. Our empirical analysis reveals that RL-trained models consistently outperform base models across a wide range of pass@$k$ evaluations, including scenarios where base models fail entirely regardless of the number of attempts. We further show that reasoning boundary improvements correlates strongly with task competence of base model and training duration, suggesting that RL can explore and populate new regions of solution space over time. These findings offer new insights into the conditions under which RL meaningfully expands reasoning boundaries in language models and establish a foundation for future work on long-horizon RL for reasoning. We will release model weights and data to support further research.

Many real-world applications require decision-making where the environmental dynamics evolve over time. These non-stationary environments pose significant challenges to traditional decision-making models, which typically assume stationary dynamics. Non-stationary Markov decision processes (NS-MDPs) offer a framework to model and solve decision problems under such changing conditions. However, there are no standardized simulation frameworks for NS-MDPs, as opposed to widely popular frameworks for stationary problems. We present NS-Gym, the first simulation toolkit designed explicitly for NS-MDPs, integrated within the popular Gymnasium framework. In NS-Gym, we segregate the evolution of the environmental parameters that characterize non-stationarity from the agent’s decision-making module, allowing for modular and flexible adaptations to dynamic environments. We review prior work in this domain and present a toolkit encapsulating key problem characteristics and types in NS-MDPs. This toolkit is the first effort to develop a set of standardized interfaces and benchmark problems to enable consistent and reproducible evaluation of algorithms under non-stationary conditions. We also benchmark several algorithmic approaches from prior work on NS-MDPs using NS-Gym. We envision that NS-Gym will enable researchers to study decision-making under non-stationarity by providing standardized interfaces, simulation frameworks, and benchmark problems.

While Multimodal Large Language Models (MLLMs) have advanced GUI navigation agents, current approaches face limitations in cross-domain generalization and effective history utilization. We present a reasoning-enhanced framework that systematically integrates structured reasoning, action prediction, and history summarization. The structured reasoning component generates coherent Chain-of-Thought analyses combining progress estimation and decision reasoning, which inform both immediate action predictions and compact history summaries for future steps. Based on this framework, we train a GUI agent, GUI-Rise, through supervised fine-tuning on pseudo-labeled trajectories and reinforcement learning with Group Relative Policy Optimization (GRPO). This framework employs specialized rewards, including a history-aware objective, directly linking summary quality to subsequent action performance. Comprehensive evaluations on standard benchmarks demonstrate state-of-the-art results under identical training data conditions, with particularly strong performance in out-of-domain scenarios. These findings validate our framework's ability to maintain robust reasoning and generalization across diverse GUI navigation tasks.