ICLR 2026 论文列表

Being able to cooperate with diverse humans is an important component of many economically valuable AI tasks, from household robotics to autonomous driving. However, generalizing to novel humans requires training on data that captures the diversity of human behaviors. Adversarial training is a promising method that allows dynamic data generation and ensures that agents are robust. It creates a feedback loop where the agent’s performance influences the generation of new adversarial data, which can be used immediately to train the agent. However, adversarial training is difficult to apply in a cooperative task; how can we train an adversarial cooperator? We propose a novel strategy that combines a pre-trained generative model to simulate valid cooperative agent policies with adversarial training to maximize regret. We call our method \textbf{GOAT}: \textbf{G}enerative \textbf{O}nline \textbf{A}dversarial \textbf{T}raining. In this framework, the GOAT dynamically searches the latent space of the generative model for coordination strategies where the learning policy---the Cooperator agent---underperforms. GOAT enables better generalization by exposing the Cooperator to various challenging interaction scenarios. We maintain realistic coordination strategies by keeping the generative model frozen, thus avoiding adversarial exploitation. We evaluate GOAT with real human partners, and the results demonstrate state-of-the-art performance on the Overcooked benchmark, highlighting its effectiveness in generalizing to diverse human behaviors.

We study the problem of offline preference-based reinforcement learning (PbRL), where the agent learns from pre-collected preference data by comparing trajectory pairs. While prior work has established theoretical foundations for offline PbRL, existing algorithms face significant practical limitations: some rely on computationally intractable optimization procedures, while others suffer from unstable training and high performance variance. To address these challenges, we propose Preference-based Value Optimization (PVO), a simple and practical algorithm that achieves both strong empirical performance and theoretical guarantees. PVO directly optimizes the value function consistent with preference feedback by minimizing a novel \emph{value alignment loss}. We prove that PVO attains a rate-optimal sample complexity of $\mathcal{O}(\varepsilon^{-2})$, and further show that the value alignment loss is applicable not only to value-based methods but also to actor–critic algorithms. Empirically, PVO achieves robust and stable performance across diverse continuous control benchmarks. It consistently outperforms strong baselines, including methods without theoretical guarantees, while requiring no additional hyperparameters for preference learning. Moreover, our ablation study demonstrates that substituting the standard TD loss with the value alignment loss substantially improves learning from preference data, confirming its effectiveness for PbRL.

Memory is crucial for enabling agents to tackle complex tasks with temporal and spatial dependencies. While many reinforcement learning (RL) algorithms incorporate memory, the field lacks a universal benchmark to assess an agent's memory capabilities across diverse scenarios. This gap is particularly evident in tabletop robotic manipulation, where memory is essential for solving tasks with partial observability and ensuring robust performance, yet no standardized benchmarks exist. To address this, we introduce MIKASA (Memory-Intensive Skills Assessment Suite for Agents), a comprehensive benchmark for memory RL, with three key contributions: (1) we propose a comprehensive classification framework for memory-intensive RL tasks, (2) we collect MIKASA-Base -- a unified benchmark that enables systematic evaluation of memory-enhanced agents across diverse scenarios, and (3) we develop MIKASA-Robo -- a novel benchmark of 32 carefully designed memory-intensive tasks that assess memory capabilities in tabletop robotic manipulation. Our work introduces a unified framework to advance memory RL research, enabling more robust systems for real-world use.

In the realm of multi-agent systems, the challenge of partial observability is a critical barrier to effective coordination and decision-making. Existing approaches, such as belief state estimation and inter-agent communication, often fall short. Belief-based methods are limited by their focus on past experiences without fully leveraging global information, while communication methods often lack a robust model to effectively utilize the auxiliary information they provide. To solve this issue, we propose Global State Diffusion Algorithm to infer the global state based on the local observations. By formulating the state inference process as a multi-modal diffusion process, GlobeDiff overcomes ambiguities in state estimation while simultaneously inferring the global state with high fidelity. We prove that the estimation error of GlobeDiff under both unimodal and multi-modal distributions can be bounded. Extensive experimental results demonstrate that GlobeDiff achieves superior performance and is capable of accurately inferring the global state.

Reinforcement learning with verifiable rewards (RLVR) has significantly improved reasoning in large language models (LLMs), yet the token-level mechanisms through which they reshape model behavior remain unclear. We present a systematic empirical study of RLVR’s distributional effects across three complementary axes: (1) token-level distributional shifts, (2) functional validation via cross-sampling interventions, and (3) exploratory investigations of advantage signal modulation based on token divergence. We find that RL fine-tuning induces sparse, targeted changes, with only a small fraction of tokens exhibiting significant distributional divergence, and we further analyze the nature of these shifts. These divergent distributions are not uniformly predicted by entropy, indicating that RLVR can modify both initially high and low entropy distributions under different settings. Cross-sampling experiments reveal that inserting just a small fraction of RL-sampled tokens into base model generations recovers most RL performance gains, while injecting a small portion of base-sampled tokens into RL generations collapses performance to base levels, functionally isolating the critical role of divergent tokens. Finally, we explore divergence-weighted variants of the advantage signal, finding that they can amplify improvements in baselines. Our work sheds light on the distributional changes induced by RLVR and provides a granular, token-level lens for understanding and improving RL fine-tuning in LLMs.

Reinforcement learning with verifiable rewards has significantly advanced reasoning with large language models (LLMs) in domains such as mathematics and logic. However, verifiable signals provide only coarse-grained or binary correctness feedback. This limitation results in inefficiencies like overly verbose or repetitive reasoning. Existing length-based solutions (e.g., length penalty) compromise accuracy. To address this deficiency, we introduce **self-aligned reward (SAR)**, a generic, universally applicable self-guided signal that complements verifiable rewards to enhance both reasoning accuracy and efficiency in RL. Specifically, SAR is defined as the relative perplexity difference between an answer conditioned on the query and the standalone answer, thereby favoring responses that are concise and query-specific. Quantitative analysis reveals that SAR reliably judges answer quality: concise, correct answers score higher than redundant ones, and partially correct answers score higher than entirely incorrect ones. Evaluation on 4 different models across 7 benchmarks shows that integrating SAR with prevalent RL algorithms like PPO and GRPO reduces answer length by 30%, while improving accuracy by 4%. Our analysis also shows that SAR generalizes well to out-of-domain tasks and achieves a Pareto-optimal frontier between correctness and efficiency compared to state-of-the-art baselines. We also show that SAR shortens unnecessary elaboration while preserving advanced reasoning behaviors. These results highlight the promise of self-aligned reward as a fine-grained complement to verifiable rewards, paving the way for efficient and effective LLM training. All of our code implementations and data are publicly available at GitHub.

Progress in long-context reasoning for large language models (LLMs) has lagged behind other recent advances. This gap arises not only from the intrinsic difficulty of processing long texts, but also from the scarcity of reliable human annotations and programmatically verifiable reward signals. In this paper, we propose SPELL, a multi-role self-play reinforcement learning framework that enables scalable, label-free optimization for long-context reasoning. SPELL integrates three cyclical roles—questioner, responder, and verifier—within a single model to enable continual self-improvement. The questioner generates questions from raw documents paired with reference answers; the responder learns to solve these questions based on the documents; and the verifier evaluates semantic equivalence between the responder’s output and the questioner's reference answer, producing reward signals to guide continual training. To stabilize training, we introduce an automated curriculum that gradually increases document length and a reward function that adapts question difficulty to the model’s evolving capabilities. Extensive experiments on six long-context benchmarks show that SPELL consistently improves performance across diverse LLMs and outperforms equally sized models fine-tuned on large-scale annotated data. Notably, SPELL achieves an average 7.6-point gain in pass@8 on the strong reasoning model Qwen3-30B-A3B-Thinking, raising its performance ceiling and showing promise for scaling to even more capable models. Our code is available at https://github.com/Tongyi-Zhiwen/Qwen-Doc.

Offline reinforcement learning enables policy learning solely from fixed datasets, without costly or risky environment interactions, making it highly valuable for real-world applications. While Transformer-based approaches have recently demonstrated strong sequence modeling capabilities, they typically learn from complete trajectories conditioned on final returns. To mitigate this limitation, we propose the Peak-Return Greedy Slicing (PRGS) framework, which explicitly partitions trajectories at the timestep level and emphasizes high-quality subtrajectories. PRGS first leverages an MMD-based return estimator to characterize the distribution of future returns for state-action pairs, yielding optimistic return estimates. It then performs greedy slicing to extract high-quality subtrajectories for training. During evaluation, an adaptive history truncation mechanism is introduced to align the inference process with the training procedure. Extensive experiments across multiple benchmark datasets indicate that PRGS significantly improves the performance of Transformer-based offline reinforcement learning methods by effectively enhancing their ability to exploit and recombine valuable subtrajectories.

The performance of reinforcement learning (RL) agents depends critically on the quality of the underlying feature representations. Hyperbolic feature spaces are well-suited for this purpose, as they naturally capture hierarchical and relational structure often present in complex RL environments. However, leveraging these spaces commonly faces optimization challenges due to the nonstationarity of RL. In this work, we identify key factors that determine the success and failure of training hyperbolic deep RL agents. By analyzing the gradients of core operations in the Poincaré ball and hyperboloid models of hyperbolic geometry, we show that large-norm embeddings destabilize gradient-based training, leading to trust-region violations in proximal policy optimization (PPO). Based on these insights, we introduce Hyper++, a new hyperbolic PPO agent that consists of three components: (1) stable critic training through a categorical value loss instead of regression; (2) feature regularization guaranteeing bounded norms while avoiding the curse of dimensionality from clipping; and (3) using a more optimization-friendly formulation of hyperbolic network layers. In experiments on ProcGen, we show that Hyper++ guarantees stable learning, outperforms prior hyperbolic agents, and reduces wall-clock time by approximately 30%. On Atari-5 with Double DQN, Hyper++ strongly outperforms Euclidean and hyperbolic baselines.

Reward design is of great importance for solving complex tasks with reinforcement learning. Recent studies have explored using image-text similarity produced by vision-language models (VLMs) to augment rewards of a task with visual feedback. A common practice linearly adds VLM scores to task or success rewards without explicit shaping, potentially altering the optimal policy. Moreover, such approaches, often relying on single static images, struggle with tasks whose desired behavior involves complex, dynamic motions spanning multiple visually different states. Furthermore, single viewpoints can occlude critical aspects of an agent's behavior. To address these issues, this paper presents Multi-View Video Reward Shaping (MVR), a framework that models the relevance of states regarding the target task using videos captured from multiple viewpoints. MVR leverages video-text similarity from a frozen pre-trained VLM to learn a state relevance function that mitigates the bias towards specific static poses inherent in image-based methods. Additionally, we introduce a state-dependent reward shaping formulation that integrates task-specific rewards and VLM-based guidance, automatically reducing the influence of VLM guidance once the desired motion pattern is achieved. We confirm the efficacy of the proposed framework with extensive experiments on challenging humanoid locomotion tasks from HumanoidBench and manipulation tasks from MetaWorld, verifying the design choices through ablation studies.

Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective for enhancing Large Language Models (LLMs) on complex reasoning tasks. Yet current methods face an exploration dilemma: standard RL struggles to escape the local optima of pre-trained LLMs’ sharply peaked initial policies, boosting single-solution accuracy (pass@1) but suppressing solution diversity and multi-solution performance (pass@k). As a result, RLVR often distills existing capabilities rather than discovering new reasoning strategies. We address this with a Risk-Sensitive Reinforcement Learning framework. By adopting a risk-seeking objective that interpolates between mean and maximum rewards, we derive a novel Risk-Sensitive GRPO (RS-GRPO) algorithm that emphasizes hard prompts to drive exploration. Across six mathematical reasoning benchmarks and five LLMs, RS-GRPO consistently improves pass@k performance while enhancing or maintaing pass@1.

We study the problem of training a risk-sensitive reinforcement learning (RL) agent through imitation learning (IL). Unlike standard IL, our goal is not only to train an agent that matches the expert’s expected return (i.e., its *average performance*) but also its *risk attitude* (i.e., other features of the return distribution, such as variance). We propose a general formulation of the risk-sensitive IL problem in which the objective is to match the expert’s return distribution in Wasserstein distance. We focus on the tabular setting and assume the expert’s reward is *known*. After demonstrating the limited expressivity of Markovian policies for this task, we introduce an efficient and sufficiently expressive subclass of non-Markovian policies tailored to it. Building on this subclass, we develop two provably efficient algorithms—RS-BC and RS-KT —for solving the problem when the transition model is unknown and known, respectively. We show that RS-KT achieves substantially lower sample complexity than RS-BC by exploiting dynamics information. We further demonstrate the sample efficiency of return distribution matching in the setting where the expert’s reward is *unknown* by designing an oracle-based variant of RS-KT. Finally, we complement our theoretical analysis of RS-KT and RS-BC with numerical simulations, highlighting both their sample efficiency and the advantages of non-Markovian policies over standard sample-efficient IL algorithms.

Safe exploration is a prerequisite for deploying reinforcement learning (RL) agents in safety-critical domains. In this paper, we approach safe exploration through the lens of epistemic uncertainty, where the actor’s sensitivity to parameter perturbations serves as a practical proxy for regions of high uncertainty. We propose Sharpness-Aware Policy Optimization (SHAPO), a sharpness-aware policy update rule that evaluates gradients at perturbed parameters, making policy updates pessimistic with respect to the actor’s epistemic uncertainty. Analytically we show that this adjustment implicitly reweighs policy gradients, amplifying the influence of rare unsafe actions while tempering contributions from already safe ones, thereby biasing learning toward conservative behavior in under-explored regions. Across several continuous-control tasks, our method consistently improves both safety and task performance over existing baselines, significantly expanding their Pareto frontiers.

We study off-dynamics offline reinforcement learning, where the goal is to learn a policy from offline source and limited target datasets with mismatched dynamics. Existing methods either penalize the reward or discard source transitions occurring in parts of the transition space with high dynamics shift. As a result, they optimize the policy using data from low-shift regions, limiting exploration of high-reward states in the target domain that do not fall within these regions. Consequently, such methods often fail when the dynamics shift is significant or the optimal trajectories lie outside the low-shift regions. To overcome this limitation, we propose MOBODY, a Model-Based Off-Dynamics Offline RL algorithm that optimizes a policy using learned target dynamics transitions to explore the target domain, rather than only being trained with the low dynamics-shift transitions. For the dynamics learning, built on the observation that achieving the same next state requires taking different actions in different domains, MOBODY employs separate action encoders for each domain to encode different actions to the shared latent space while sharing a unified representation of states and a common transition function. We further introduce a target Q-weighted behavior cloning loss in policy optimization to avoid out-of-distribution actions, which push the policy toward actions with high target-domain Q-values, rather than high source domain Q-values or uniformly imitating all actions in the offline dataset. We evaluate MOBODY on a wide range of MuJoCo and Adroit benchmarks, demonstrating that it outperforms state-of-the-art off-dynamics RL baselines as well as policy learning methods based on different dynamics learning baselines, with especially pronounced improvements in challenging scenarios where existing methods struggle.

Existing reinforcement learning (RL) methods struggle with complex dynamical systems that demand interactions at high frequencies or irregular time intervals. Continuous-time RL (CTRL) has emerged as a promising alternative by replacing discrete-time Bellman recursion with differentiable value functions defined as viscosity solutions of the Hamilton–Jacobi–Bellman (HJB) equation. While CTRL has shown promise, its applications have been largely limited to the single-agent domain. This limitation stems from two key challenges: (i) conventional methods for solving HJB equations suffer from the curse of dimensionality (CoD), making them intractable in high-dimensional systems; and (ii) even with learning-based approaches to alleviate the CoD, accurately approximating centralized value functions in multi-agent settings remains difficult, which in turn destabilizes policy training. In this paper, we propose a CT-MARL framework that uses physics-informed neural networks (PINNs) to approximate HJB-based value functions at scale. To ensure the value is consistent with its differential structure, we align value learning with value-gradient learning by introducing a Value Gradient Iteration (VGI) module that iteratively refines value gradients along trajectories. This improves gradient accuracy, in turn yielding more precise value approximations and stronger policy learning. We evaluate our method using continuous‑time variants of standard benchmarks, including multi‑agent particle environment (MPE) and multi‑agent MuJoCo. Our results demonstrate that our approach consistently outperforms existing continuous‑time RL baselines and scales to complex cooperative multi-agent dynamics.

Generative models have recently demonstrated remarkable success across diverse domains, motivating their adoption as expressive policies in reinforcement learning (RL). While they have shown strong performance in offline RL, particularly where the target distribution is well defined, their extension to online fine-tuning has largely been treated as a direct continuation of offline pre-training, leaving key challenges unaddressed. In this paper, we propose Flow Matching with Injected Noise for Offline-to-Online RL (FINO), a novel method that leverages flow matching-based policies to enhance sample efficiency for offline-to-online RL. FINO facilitates effective exploration by injecting noise into policy training, thereby encouraging a broader range of actions beyond those observed in the offline dataset. In addition to exploration-enhanced flow policy training, we combine an entropy-guided sampling mechanism to balance exploration and exploitation, allowing the policy to adapt its behavior throughout online fine-tuning. Experiments across diverse, challenging tasks demonstrate that FINO consistently achieves superior performance under limited online budgets.

Model fusion has emerged as a promising research direction, offering a resource-efficient paradigm that leverages existing pre-trained models to circumvent the need for training from scratch. In this work, we investigate the fusion of models specifically adapted for decision-making tasks. This challenge divides into two distinct, yet related subproblems: the direct fusion of models that act as policy and the fusion of dynamics models that subsequently induce a policy. We suggest that these seemingly divergent subproblems can be unified through the lens of energy-based models (EBMs), which parameterizes a conditional distribution via an energy function where lower energy implies higher probability. Our framework, \textbf{EMFuse}, provides this convergence by leveraging the concept of energy as a common currency for fusion. For direct fusion of policies, such as those in language models, the output distribution is commonly softmax (Boltzmann), which essentially defines the negative logarithmic probability as an energy function. For dynamics models, existing works often train a set of models on the same dataset to obtain robust uncertainty estimation; such an ensemble approach leads to an exponential explosion in computational complexity when it comes to dynamics fusion across multiple sets of models. To overcome this, we introduce the Any-step Dynamics Energy-based Transition Model (ADETM), a novel architecture that performs efficient single-model-per-dataset uncertainty estimation with its energy-based backbone, thereby avoiding this computational explosion. Our EMFuse framework surpasses other baselines by 0.34\% to 6.63\% on single/cross domain discrete decision-making benchmarks, and achieved an extra 2.3 to 7.4 normalized points on average in D4RL MuJoCo continuous-control scenarios.

Despite the significant advances in Deep Reinforcement Learning (RL) observed in the last decade, the amount of training experience necessary to learn effective policies remains one of the primary concerns in both simulated and real environments. Looking to solve this issue, previous work has shown that improved efficiency can be achieved by separately modeling the agent and environment, but usually requires a supervisory signal. In contrast to RL, humans can perfect a new skill from a small number of trials and often do so without a supervisory signal, making neuroscientific studies of human development a valuable source of inspiration for RL. In particular, we explore the idea of motor prediction, which states that humans develop an internal model of themselves and of the consequences that their motor commands have on the immediate sensory inputs. Our insight is that the movement of the agent provides a cue that allows the duality between the agent and environment to be learned. To instantiate this idea, we present Ego-Foresight (EF), a self-supervised method for disentangling agent information based on motion and prediction. Our main finding is that, when used as an auxiliary task in feature learning, self-supervised agent-awareness improves the sample-efficiency and performance of the underlying RL algorithm. To test our approach, we study the ability of EF to predict agent movement and disentangle agent information. Then, we integrate EF with both model-free and model-based RL algorithms to solve simulated control tasks, showing improved sample-efficiency and performance.

Gradient-regularized value learning methods improve sample efficiency by leveraging learned models of transition dynamics and rewards to estimate return gradients. However, existing approaches, such as MAGE, struggle in stochastic or noisy environments, limiting their applicability. In this work, we address these limitations by extending distributional reinforcement learning on continuous state-action spaces to model not only the distribution over scalar state-action value functions but also over their gradients. We refer to this approach as Distributional Sobolev Training. Inspired by Stochastic Value Gradients (SVG), our method utilizes a one-step world model of reward and transition distributions implemented via a conditional Variational Autoencoder (cVAE). The proposed framework is sample-based and employs Max-sliced Maximum Mean Discrepancy (MSMMD) to instantiate the distributional Bellman operator. We prove that the Sobolev-augmented Bellman operator is a contraction with a unique fixed point, and highlight a fundamental smoothness trade-off underlying contraction in gradient-aware RL. To validate our method, we first showcase its effectiveness on a simple stochastic reinforcement‐learning toy problem, then benchmark its performance on several MuJoCo environments.