ICLR 2026 Papers

Large-scale reinforcement learning with verifiable rewards (RLVR) has proven effective in harnessing the potential of large language models (LLMs) for single-turn reasoning tasks. In realistic reasoning scenarios, LLMs often rely on external tools to assist in task-solving processes. However, current RL algorithms typically employ trajectory-level rollout sampling, consistently neglecting the fine-grained exploration of multi-turn tool-call steps. To bridge this gap, we propose Agentic Reinforced Policy Optimization (ARPO), a novel agentic RL algorithm tailored for training multi-turn LLM-based agents. Our preliminary experiments reveal that LLMs frequently exhibit increased uncertainty after tool-call steps, as evidenced by higher entropy in the distribution of generated tokens. Motivated by this, ARPO incorporates an entropy-based adaptive rollout mechanism, encouraging the policy model to adaptively branch sampling during high-entropy tool-call rounds, thereby promoting step-level exploration of latent tool-use behaviors. By integrating an advantage attribution estimation, ARPO enables LLMs to internalize advantage differences in stepwise tool-use interactions. Experiments across 13 challenging benchmarks demonstrate ARPO's superiority over trajectory-level RL algorithms. Remarkably, ARPO achieves improved performance using only half of the tool-use budget required by existing methods, offering a scalable solution for aligning LLM-based agents with real-time dynamic environments. Our codes are released at https://github.com/RUC-NLPIR/ARPO.

Vision-Language-Action (VLA) models have emerged as a powerful paradigm for robotic manipulation. Despite substantial progress enabled by large-scale pretraining and supervised fine-tuning (SFT), these models face two fundamental challenges: (i) the scarcity and high cost of large-scale robotic trajectories required for SFT scaling, and (ii) limited generalization to tasks under distribution shift. To overcome these limitations, we explore reinforcement learning (RL) as a pathway to scaling VLA training beyond limited datasets. Inspired by LLM breakthroughs where RL with outcome rewards enhances step-by-step reasoning, we ask: Can outcome-driven RL improve long-horizon step-by-step action planning of VLA? In this work, we introduce SimpleVLA-RL, an efficient RL framework tailored for VLA models. Building upon veRL, we introduce VLA-specific trajectory sampling, scalable parallelization, multi-environment rendering, and optimized loss computation. Applied to OpenVLA-OFT, SimpleVLA-RL achieves 99\% of SoTA performance on LIBERO and 80\% relative improvement on RoboTwin 1.0\&2.0, outperforming $\pi_0$ with our proposed exploration-enhancing strategies. SimpleVLA-RL reduces dependence on large-scale data, enables robust generalization, and remarkably surpasses SFT in real-world tasks. Moreover, we identify a novel phenomenon "pushcut'' during RL training, wherein the policy discovers unseen patterns beyond those seen in previous training process.

Out-of-distribution (OOD) detection is essential for reliable deployment of machine learning systems in vision, robotics, and reinforcement learning. We introduce Score–Curvature Out-of-distribution Proximity Evaluator for Diffusion (SCOPED), a fast and general-purpose OOD detection method for diffusion models that reduces the number of forward passes on the trained model by an order of magnitude compared to prior methods, outperforming most diffusion-based baselines and approaching the accuracy of the strongest ones. SCOPED is computed from a single diffusion model trained once on a diverse dataset and combines the Jacobian trace and squared norm of the model’s score function into a single test statistic. Rather than thresholding on a fixed value, we estimate the in-distribution density of SCOPED scores using kernel density estimation, enabling a flexible, unsupervised test that, in the simplest case, only requires a single forward pass and one Jacobian–vector product (JVP), made efficient by Hutchinson’s trace estimator. On four vision benchmarks, SCOPED achieves competitive or state-of-the-art precision-recall scores despite its low computational cost. The same method generalizes to robotic control tasks with shared state and action spaces, identifying distribution shifts across reward functions and training regimes. These results position SCOPED as a practical foundation for fast and reliable OOD detection in real-world domains, including perceptual artifacts in vision, outlier detection in autoregressive models, exploration in reinforcement learning, and dataset curation for unsupervised training.

Recent Large Language Models (LLMs) have made remarkable progress, but they still struggle with complex reasoning tasks such as logical deduction and planning. This is partly because they rely primarily on token-level probability relationships, which limits their ability to reason effectively. In this paper, inspired by cognitive science and neurosymbolic AI, we introduce Structured Reasoning, which aimes at enhancing the reasoning capabilities of LLMs from the step level. To this end, we first collect high‑frequency, domain‑agnostic reasoning step tags and construct a structured reasoning dataset with those tags. Then, we treat a reasoning process as a directed acyclic graph, where the vertices represent steps and the edges indicate the direction of reasoning. In this context, an efficient reasoning process corresponds to, or can be characterized by, a sparse reasoning graph. To construct reasoning graphs, we introduce structured tags for reliable step extraction from LLM outputs. For single-graph optimization, we propose the MaxFlow reward, which rewards graphs with balanced node contributions and fewer redundant steps. The quality of a sparse reasoning graph can be reflected by the total flow from all steps to the final answer. For multi-graph comparison, we propose the LCS reward, which selects reliable reasoning paths by identifying optimal common subsequences (consecutive steps) shared across multiple generated responses (sequences). Experiments with DeepSeek-R1-Distill-Qwen-1.5B and 7B models show that our method consistently outperforms GRPO and other carefully tuned baselines across various context lengths (0.5k–8k). Structured Reasoning shows particular strength in efficiency (better performance with fewer steps) and stability (consistently generating high-quality outputs across a temperature range of 0.1 to 1.0).

Most existing offline RL methods presume the availability of action labels within the dataset, but in many practical scenarios, actions may be missing due to privacy, storage, or sensor limitations. We formalise the setting of action-free offline-to-online RL, where agents must learn from datasets consisting solely of $(s,r,s')$ tuples and later leverage this knowledge during online interaction. To address this challenge, we propose learning state policies that recommend desirable next-state transitions rather than actions. Our contributions are twofold. First, we introduce a simple yet novel state discretisation transformation and propose Offline State-Only DecQN (OSO-DecQN), a value-based algorithm designed to pre-train state policies from action-free data. OSO-DecQN integrates the transformation to scale efficiently to high-dimensional problems while avoiding instability and overfitting associated with continuous state prediction. Second, we propose a novel mechanism for guided online learning that leverages these pre-trained state policies to accelerate the learning of online agents. Together, these components establish a scalable and practical framework for leveraging action-free datasets to accelerate online RL. Empirical results across diverse benchmarks demonstrate that our approach improves convergence speed and asymptotic performance, while analyses reveal that discretisation and regularisation are critical to its effectiveness.

The automation of analog integrated circuit (IC) design remains a longstanding challenge, primarily due to the intricate interdependencies among physical layout, parasitic effects, and circuit-level performance. These interactions impose complex constraints that are difficult to accurately capture and optimize using conventional design methodologies. Although recent advances in machine learning (ML) have shown promise in automating specific stages of the analog design flow, the development of holistic, end-to-end frameworks that integrate these stages and iteratively refine layouts using post-layout, parasitic-aware performance feedback is still in its early stages. Furthermore, progress in this direction is hindered by the limited availability of open, high-quality datasets tailored to the analog domain, restricting both the benchmarking and the generalizability of ML-based techniques. To address these limitations, we present OSIRIS, a scalable dataset generation pipeline for analog IC design. OSIRIS systematically explores the design space of analog circuits while producing comprehensive performance metrics and metadata, thereby enabling ML-driven research in electronic design automation (EDA). In addition, we release a dataset consisting of 87,100 circuit variations generated with OSIRIS, accompanied by a reinforcement learning (RL)–based baseline method that exploits OSIRIS for analog design optimization.

Knowledge-Based Visual Question Answering (KB-VQA) requires models to answer questions about an image by integrating external knowledge, posing significant challenges due to noisy retrieval and the structured, encyclopedic nature of the knowledge base. These characteristics create a distributional gap from pretrained multimodal large language models (MLLMs), making effective reasoning and domain adaptation difficult in the post-training stage. In this work, we propose \textit{Wiki-R1}, a data-generation-based curriculum reinforcement learning framework that systematically incentivizes reasoning in MLLMs for KB-VQA. Wiki-R1 constructs a sequence of training distributions aligned with the model’s evolving capability, bridging the gap from pretraining to the KB-VQA target distribution. We introduce \textit{controllable curriculum data generation}, which manipulates the retriever to produce samples at desired difficulty levels, and a \textit{curriculum sampling strategy} that selects informative samples likely to yield non-zero advantages during RL updates. Sample difficulty is estimated using observed rewards and propagated to unobserved samples to guide learning. Experiments on two KB-VQA benchmarks, Encyclopedic VQA and InfoSeek, demonstrate that Wiki-R1 achieves new state-of-the-art results, improving accuracy from 35.5\% to 37.1\% on Encyclopedic VQA and from 40.1\% to 44.1\% on InfoSeek.

A/B testing has become a gold standard for modern technological companies to conduct policy evaluation. Yet, its application to time series experiments, where treatments are sequentially assigned over time, remains challenging. Existing designs suffer from two limitations: (i) they do not fully leverage the entire history for treatment allocation; (ii) they rely on strong assumptions to approximate the objective function (e.g., the mean squared error of the estimated treatment effect) for optimizing the design. We first establish an impossibility theorem showing that failure to condition on the full history leads to suboptimal designs, due to the dynamic dependencies in time series experiments. To address both limitations simultaneously, we next propose a transformer reinforcement learning (RL) approach which leverages transformers to condition treatment allocation on the entire history and employs RL to directly optimize the MSE without relying on restrictive assumptions. Empirical evaluations on synthetic data, a publicly available dispatch simulator, and a real-world ridesharing dataset demonstrate that our proposal consistently outperforms existing designs.

Dynamic programming (DP) is central to combinatorial optimization, optimal control, and reinforcement learning, yet its perceived sequentiality has long hindered scalability. We introduce a general-purpose GPU framework that reformulates broad classes of forward DP recursions as batched min--plus matrix--vector products over layered DAGs, collapsing actions into masked state-to-state transitions that map directly to GPU kernels. This approach removes a major bottleneck in scenario-based stochastic programming (SP), where the use of DP has traditionally restricted the number of scenarios due to excessive computational cost. Our framework exposes massive parallelism across scenarios, transition layers, and, when applicable, route or action options, via self-designed GPU kernels that implement Bellman updates with warp-/block-level reductions and numerically safe masking. In a single GPU pass, these kernels can process over $10^6$ uncertainty realizations, far beyond the capacity of prior scenario-based methods. We demonstrate the approach in two canonical SP applications: (i) a vectorized split operator for the capacitated vehicle routing problem with stochastic demand, exploiting **2D** parallelism (scenarios $\times$ transitions); and (ii) a forward inventory reinsertion DP under an order-up-to policy, exploiting **3D** parallelism (scenarios $\times$ inventory transitions $\times$ route options). Across benchmarks, the implementation scales nearly linearly in the number of scenarios and achieves one to three orders of magnitude speedups over multithreaded CPU baselines, yielding tighter SAA estimates and consistently stronger first-stage decisions under identical wall-clock budgets. Viewed as hardware-aware software primitives, our min--plus DP kernels offer a drop-in path to scalable, GPU-accelerated stochastic discrete optimization.

Group-based reinforcement learning (RL), such as GRPO, has advanced the capabilities of large language models on long-horizon agentic tasks. To enable more fine-grained policy updates, recent research has increasingly shifted toward stepwise group-based policy optimization, which treats each step in a rollout trajectory independently while using a memory module to retain historical context. However, we find a key issue in estimating stepwise relative advantages, namely context inconsistency, where steps within the same group may differ in their historical contexts. Empirically, we reveal that this issue can lead to severely biased advantage estimation, thereby degrading policy optimization significantly. To address the issue, in this paper, we propose Hierarchical-of-Groups Policy Optimization (HGPO) for long-horizon agentic tasks. Specifically, within a group of rollout trajectories, HGPO assigns each step to multiple hierarchical groups according to the consistency of historic contexts. Then, for each step, HGPO computes distinct advantages within each group and aggregates them with an adaptive weighting scheme. In this way, HGPO can achieve a favorable bias-variance trade-off in stepwise advantage estimation, without extra models or rollouts. Evaluations on two challenging agentic tasks, ALFWorld and WebShop with Qwen2.5-1.5B-Instruct and Qwen2.5-7B-Instruct, show that HGPO significantly outperforms existing agentic RL methods under the same computational constraints.

Latent prediction–where agents learn by predicting their own latents–has emerged as a powerful paradigm for training general representations in machine learning. In reinforcement learning (RL), this approach has been explored to define auxiliary losses for a variety of settings, including reward-based and unsupervised RL, behavior cloning, and world modeling. While existing methods are typically limited to single-task learning, one-step prediction, or on-policy trajectory data, we show that temporal difference (TD) learning enables learning representations predictive of long-term latent dynamics across multiple policies from offline, reward-free transitions. Building on this, we introduce TD-JEPA, which leverages TD-based latent-predictive representations into unsupervised RL. TD-JEPA trains explicit state and task encoders, a policy-conditioned multi-step predictor, and a set of parameterized policies directly in latent space. This enables zero-shot optimization of any reward function at test time. Theoretically, we show that an idealized variant of TD-JEPA avoids collapse with proper initialization, and learns encoders that capture a low-rank factorization of long-term policy dynamics, while the predictor recovers their successor features in latent space. Empirically, TD-JEPA matches or outperforms state-of-the-art baselines on locomotion, navigation, and manipulation tasks across 13 datasets in ExoRL and OGBench, especially in the challenging setting of zero-shot RL from pixels.

Large Language Models (LLMs) show remarkable code generation capabilities but often produce imperfect code with subtle bugs. A critical bottleneck for improving code quality is the scarcity of high-quality test cases. Existing approaches, primarily based on Supervised Fine-Tuning (SFT) over static datasets, are limited in their ability to discover novel bugs and struggle with the fundamental trade-off between generating error-triggering inputs and maintaining correct expected outputs. To address these limitations, we reframe test case generation as an iterative, adversarial process. We introduce ATGEN (Adversarial Test Generator), a novel framework that trains a test case generator via Reinforcement Learning (RL) in an adversarial loop with an evolving code generator. Instead of learning from a fixed set of bugs, our test generator is dynamically trained to create "attacking" I/O pairs for buggy code that is itself being iteratively generated. This process is guided by a reward function that explicitly balances the dual objectives of maximizing the bug detection rate and maintaining high output accuracy. Extensive experiments show that ATGEN dramatically outperforms the state-of-the-art SFT-based approach, UTGen, improving IO Accuracy by nearly 40 absolute points (71.56% vs. 31.83%) and more than doubling the Attack Rate (34.02% vs. 16.24%). The adversarial curriculum is particularly effective for hard-to-detect bugs, achieving an attack rate more than double that of the strongest baseline. Furthermore, tests generated by ATGEN serve as a more effective filter in Best-of-N code generation, significantly closing the gap to the human expert upper bound. Our work establishes a new and more effective paradigm for automated test generation and debugging for LLMs.

Robot learning increasingly relies on simulation to advance complex ability such as dexterous manipulations and precise interactions, necessitating high-quality digital assets to bridge the sim-to-real gap. However, existing open-source articulated object datasets for simulation are limited by insufficient visual realism and low physical fidelity, which hinder their utility for training models mastering robotic tasks in real world. To address these challenges, we introduce ArtVIP, a comprehensive open-source dataset comprising high-quality digital-twin articulated objects, accompanied by indoor-scene assets.Crafted by professional 3D modelers adhering to unified standards, ArtVIP ensures visual realism through precise geometric meshes and high-resolution textures, and physical fidelity is achieved via fine-tuned dynamic parameters. Meanwhile, the dataset pioneers embedded modular interaction behaviors within assets and pixel-level affordance annotations. Feature-map visualization and optical motion capture are employed to quantitatively demonstrate ArtVIP's visual and physical fidelity, with its applicability validated across imitation learning and reinforcement learning experiments. Provided in USD format with detailed production guidelines, ArtVIP is fully open-source, benefiting the research community and advancing robot learning research.

We study distributed reinforcement learning (RL) with policy gradient methods under asynchronous and parallel computations and communications. While non-distributed methods are well understood theoretically and have achieved remarkable empirical success, their distributed counterparts remain less explored, particularly in the presence of heterogeneous asynchronous computations and communication bottlenecks. We introduce two new algorithms, Rennala NIGT and Malenia NIGT, which implement asynchronous policy gradient aggregation and achieve state-of-the-art efficiency. In the homogeneous setting, Rennala NIGT provably improves the total computational and communication complexity while supporting the AllReduce operation. In the heterogeneous setting, Malenia NIGT simultaneously handles asynchronous computations and heterogeneous environments with strictly better theoretical guarantees. Our results are further corroborated by experiments, showing that our methods significantly outperform prior approaches.

Recent advances in reasoning capabilities of large language models (LLMs) are largely driven by reinforcement learning (RL), yet the underlying parameter dynamics during RL training remain poorly understood. This work identifies two fundamental properties of RL-induced parameter updates in LLMs: (1) Rank-1 Dominance, where the top singular subspace of the parameter update matrix nearly fully determines reasoning improvements, recovering over 99\% of performance gains; and (2) Rank-1 Linear Dynamics, where this dominant subspace evolves linearly throughout training, enabling accurate prediction from early checkpoints. Extensive experiments across 13 LLMs and 10 algorithms validate the generalizability of these properties. More importantly, based on these findings, we propose AlphaRL, a plug-in acceleration framework that extrapolates the final parameter update using a short early training window, achieving up to 2.5× speedup while retaining > 96\% of reasoning performance without extra modules or hyperparameter tuning. This positions our finding as a versatile and practical tool for large-scale RL, opening a path toward principled, interpretable, and efficient training paradigm for LLMs.

Reinforcement Learning (RL) in partially observable environments poses significant challenges due to the complexity of learning under uncertainty. While additional information, such as that available in simulations, can enhance training, effectively leveraging it remains an open problem. To address this, we introduce Guided Policy Optimization (GPO), a framework that co-trains a guider and a learner. The guider takes advantage of privileged information while ensuring alignment with the learner's policy that is primarily trained via imitation learning. We theoretically demonstrate that this learning scheme achieves optimality comparable to direct RL, thereby overcoming key limitations inherent in existing approaches. Empirical evaluations show strong performance of GPO across various tasks, including continuous control with partial observability and noise, and memory-based challenges, significantly outperforming existing methods.

Reinforcement Learning with Verifiable Rewards (RLVR) is a leading approach for tuning language models on mathematical reasoning tasks. However, it remains unclear whether RLVR's gains stem from genuine reasoning improvements or simply from steering the model toward answer formats that already appear in the reference distribution. Inspired by recent evidence \citep{zhao2025echo,yue2025does}, we study this question by optimizing only the first $k$ tokens (e.g. $k=32$) of each solution, generating the remainder of the response from the reference model. We study two methods for prefix optimization, using a naive algorithm that clusters prefixes and selects the best prefix (Prefix Clustering), and a method that optimizes the prefix by finetuning a lightweight adapter model with RL (Prefix-RL). We show that tuning only the first $k$ tokens can significantly improve the accuracy on math, suggesting that at least some of the gains from RL are due to upweighting a preferable solution strategy. Our results suggest that simple prefix optimization methods can provide an efficient alternative to RL, delivering substantial improvements across different models and benchmarks for a tiny fraction of the compute required for standard RL, and that these gains are robust across prefix lengths and random seeds.

Traditional RLHF optimizes language models with coarse, scalar rewards that mask the fine-grained reasons behind success or failure, leading to slow, opaque learning. Recent work augments RL with textual critiques through prompting or reflection, improving interpretability but leaving model parameters untouched. We introduce **Text2Grad**, a reinforcement-learning paradigm that *turns free-form textual feedback into span-level gradients*. Given human (or programmatic) critiques, Text2Grad aligns each feedback phrase with the relevant token spans, converts these alignments into differentiable reward signals, and performs gradient updates that directly refine the offending portions of the model's policy. This yields precise, feedback-conditioned adjustments instead of global nudges. Text2Grad is realized through three components: (1) a high-quality feedback–annotation pipeline that pairs critiques with token spans; (2) a fine-grained reward model that predicts span-level reward on answers while generating explanatory critiques; and (3) a span-level policy optimizer that back-propagates *natural-language gradients*. Across summarization, code generation, and question answering, Text2Grad consistently surpasses scalar-reward RL and prompt-only baselines, providing both higher task metrics and richer interpretability. Our results suggest that natural-language feedback can serve not only as explanations, but also as actionable training signals for fine-grained alignment. The code for our method is available at *[https://github.com/microsoft/Text2Grad](https://github.com/microsoft/Text2Grad)*.

We present DuPO, a dual learning-based preference optimization framework that generates annotation-free feedback via the generalized duality. DuPO addresses two key limitations: Reinforcement Learning with Verifiable Rewards (RLVR)’s reliance on costly labels and applicability restricted to verifiable tasks, and traditional dual learning’s restriction to strictly dual task pairs (e.g., translation and back-translation). Specifically, DuPO decomposes a primal task’s input into known and unknown components, then constructs its dual task to reconstruct the unknown part using the primal output and known information (e.g., reversing math solutions to recover hidden variables), broadening applicability to non-invertible tasks. The quality of this reconstruction serves as a self-supervised reward to optimize the primal task, synergizing with LLMs’ ability to instantiate both tasks via a single model. Empirically, DuPO achieves substantial gains across diverse tasks: it enhances the average translation quality by 2.1 COMET over 756 directions, boosts the mathematical reasoning accuracy by an average of 6.2 points on three challenge benchmarks, and enhances performance by 9.3 points as an inference-time reranker (trading computation for accuracy). These results position DuPO as a scalable, general, and annotation-free paradigm for LLM optimization.