Policy Optimization

Follow this topic Ask about this topic

Policy optimization studies how to update a decision-making rule so it earns higher rewards while meeting constraints such as safety, fairness, or limited data. Common approaches include policy gradients, actor-critic methods, trust regions and mirror descent, as well as policy iteration and newer generative-policy methods. Work spans continuous-control benchmarks such as MuJoCo, offline and multi-agent settings, and language-model training for mathematical reasoning, coding, and tool use, where token-, segment-, or trajectory-level rewards guide updates. A central challenge is balancing effective exploration and sample reuse with stable updates, useful credit assignment, and improvements that carry over from training to deployment.

Latest papers

224 papers32 papers in the last four weeks, up 52% on the four weeks before. 0.2% of all new papers.

Apr 13, 2026

DDO-RM: Distribution-Level Policy Improvement after Reward Learning

Recent theory suggests that reward-model-first methods can be more sample-efficient than direct policy fitting when the reward function is statistically simpler than the induced policy. We propose DDO-RM, a finite-candidate decision-optimization method that converts reward scores into an explicit target distribution. Unlike PPO-based RLHF or DPO, DDO-RM performs a KL-regularized mirror-descent update to project the policy toward a reward-improved distribution over a candidate set. Preliminary experiments on Pythia-410M show that DDO-RM outperforms DPO in pair accuracy (0.52 to 0.56) and mean margin (0.13 to 0.53). Our framework provides a principled connection between reward learning and mirror-descent policy improvement.

Tiantian Zhang, Jierui Zuo, Michael Chen +1

Department of Computer Science Columbia University, New York, NY 10027, USA · Department of Management Science and Engineering Tsinghua University, Beijing, China · Department of Computer Science Carnegie Mellon University, Pittsburgh, PA 15213, USA

Policy Optimization KL-Regularized RL Reward Modeling

Apr 11, 2026

Adapt to Thrive! Adaptive Power-Mean Policy Optimization for Improved LLM Reasoning

Reinforcement Learning with Verifiable Rewards (RLVR) is an essential paradigm that enhances the reasoning capabilities of Large Language Models (LLMs). However, existing methods typically rely on static policy optimization schemes that misalign with the model's evolving reasoning capabilities. To address this issue, we propose Adaptive Power-Mean Policy Optimization (APMPO), which comprises two main innovations: Power-Mean Policy Optimization (PMPO) and Feedback-Adaptive Clipping (FAC). Specifically, PMPO introduces a generalized power-mean objective. This enables the model to adaptively transition from the signal-amplifying behavior of the arithmetic mean to the consistency-enforcing behavior of the geometric mean. FAC adaptively adjusts clipping bounds based on real-time reward statistics to overcome the limitations of static mechanisms. Capitalizing on these innovations, APMPO improves learning dynamics and reasoning performance. Extensive experiments on nine datasets across three reasoning tasks showcase the superiority of APMPO over state-of-the-art RLVR-based baselines. For instance, APMPO boosts the average Pass@1 score on mathematical reasoning benchmarks by 3.0 points compared to GRPO when using Qwen2.5-3B-Instruct.

Yiming Huang, Zhenbo Shi, Shuzheng Gao +3

Harbin Institute of Technology, Shenzhen · The Chinese University of Hong Kong · Peng Cheng Laboratory

RL for Language Model Reasoning Policy Optimization LLM Reasoning Reinforcement Learning with Verifiable Rewards

Apr 9, 2026

Adaptive Simulation Experiment for LLM Policy Optimization

Large language models (LLMs) have significant potential to improve operational efficiency in operations management. Deploying these models requires specifying a policy that governs response quality, shapes user experience, and influences operational value. In this research, we treat LLMs as stochastic simulators and propose a pairwise comparison-based adaptive simulation experiment framework for identifying the optimal policy from a finite set of candidates. We consider two policy spaces: an unstructured space with no parametric assumption, and a structured space in which the data are generated from a preference model. For both settings, we characterize the fundamental data requirements for identifying the optimal policy with high probability. In the unstructured case, we derive a closed-form expression for the optimal sampling proportions, together with a clear operational interpretation. In the structured case, we formulate a regularized convex program to compute the optimal proportions. We then develop an adaptive experimental procedure, termed LLM-PO, for both policy spaces, and prove that it identifies the optimal policy with the desired statistical guarantee while asymptotically attaining the fundamental data requirements. Numerical experiments demonstrate that LLM-PO consistently outperforms benchmark methods and improves LLM performance.

Mingjie Hu, Siyang Gao, Jian-qiang Hu +1

School of Management, Fudan University · H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology · Department of Systems Engineering, City University of Hong Kong

Active Experimental Design Policy Optimization Best-Arm Identification Large Language Model-Guided Optimization Pairwise Comparison

Apr 7, 2026

Target Policy Optimization

In RL, given a prompt, we sample a group of completions from a model and score them. Two questions follow: which completions should gain probability mass, and how should the parameters move to realize that change? Standard policy-gradient methods answer both at once, so the update can overshoot or undershoot depending on the learning rate, clipping, and other optimizer choices. We introduce Target Policy Optimization (TPO), which separates the two questions. Given scored completions, TPO constructs a target distribution

q_i \propto p_i^{\,\mathrm{old}} \exp(u_i)

and fits the policy to it by cross-entropy. The loss gradient on sampled-completion logits is

p^θ- q

, which vanishes once the policy matches the target. On tabular bandits, transformer sequence tasks, and billion-parameter LLM RLVR, TPO matches PG, PPO, GRPO, and DG on easy tasks and substantially outperforms them under sparse reward. Code is available at https://github.com/JeanKaddour/tpo.

Jean Kaddour

Policy Optimization RL Post-Training RL for Language Model Reasoning

Apr 2, 2026

Posterior Optimization with Clipped Objective for Bridging Efficiency and Stability in Generative Policy Learning

Expressive generative models have advanced robotic manipulation by capturing complex, multi-modal action distributions over temporally extended trajectories. However, fine-tuning these policies via RL remains challenging due to instability and sample inefficiency. We introduce Posterior Optimization with Clipped Objective (POCO), a principled RL framework that formulates policy improvement as a posterior inference problem tailored for temporal action chunks. Through an Expectation-Maximization procedure, POCO distills a reward-weighted implicit posterior into the policy without likelihood estimation. Furthermore, POCO adopts an offline-to-online paradigm that anchors online exploration to pre-trained priors, and its model-agnostic design scales to fine-tune large VLA models without architectural modifications. Evaluations across 7 simulation benchmarks and 4 contact-rich real-world tasks demonstrate that POCO prevents catastrophic policy collapse, outperforms SOTA baselines, and achieves a 96.7% success rate on real-world tasks. Videos are available at our project website https://cccedric.github.io/poco/.

Yuhui Chen, Haoran Li, Zhennan Jiang +4

Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China · School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100049, China · CFCS, School of Computer Science, Peking University, Beijing 100871

Policy Optimization Robotic RL RL for Generative Models

Apr 1, 2026

Policy Improvement Reinforcement Learning

Reinforcement learning has become a central post-training paradigm for improving LLM and agent capabilities. Yet existing RL post-training methods share a common blind spot: they construct local learning signals from sampled trajectories, rewards, or feedback-conditioned targets, then update the policy without explicitly verifying whether the resulting policy outperforms its predecessor. Optimizing these local signals does not necessarily produce a better policy, while finite sampling, generation stochasticity and feedback noise can further widen this gap. We argue that the missing ingredient is policy improvement feedback: the ability to measure progress across policy iterations. We introduce Policy Improvement Reinforcement Learning (PIRL), which formulates inter-iteration performance gain as an explicit objective structurally aligned with final task performance. Building on PIRL, we propose Policy Improvement Policy Optimization (PIPO), a plug-in closed-loop framework that verifies the previous update against a sliding-window historical performance anchor. PIPO uses this improvement feedback to modulate the local learning signal of the base policy optimization algorithm, reinforcing updates associated with measured progress and suppressing those associated with performance drops. We provide theoretical evidence that PIPO locally aligns policy updates with the PIRL improvement objective. Experiments on mathematical reasoning, code, tool-use, and self-distillation settings show that PIPO yields consistent gains across PPO, group-relative, and self-distillation policy optimization families.

Huaiyang Wang, Xiaojie Li, Xiaohan Wang +10

Beihang University · Meituan · Peking University

RL for Language Models RL for Language Model Reasoning Policy Optimization Proximal Policy Optimization RL Post-Training Policy Learning

Mar 23, 2026

Path-Space Mirror Descent for On-Policy Reinforcement Learning under the Generalized Schrödinger Bridge

Classical on-policy algorithms such as PPO and mirror descent policy optimization provide stable proximal policy updates through tractable action likelihoods, but are typically instantiated with simple Gaussian policies whose expressiveness can be limited in complex continuous-control tasks. Generative policies based on diffusion and flow models provide more expressive action distributions, but they naturally define distributions over multi-step denoising paths whose terminal action density is often intractable, creating a mismatch with likelihood-based on-policy proximal updates. To address this mismatch, we introduce GSB-MDPO (Generalized Schrödinger Bridge Mirror Descent Policy Optimization), which formulates on-policy generative policy optimization as a Generalized Schrödinger Bridge problem over state-conditioned generation paths and instantiates the resulting path-measure update through mirror descent policy optimization. The key insight is that the GSB path-space KL plays the role of the proximal term in MDPO while upper-bounding the terminal action KL, enabling direct control of the executed action distribution without explicit terminal action likelihood evaluation. Experiments on 14 continuous-control tasks across Playground and Gym-MuJoCo demonstrate the empirical effectiveness of GSB-MDPO and support path-space regularization as a principled proximal update for multi-step generative policies.

Yuehu Gong, Zeyuan Wang, Yulin Chen +3

School of Data Science, Fudan University · Laboratory for Big Data and Decision, National University of Defense Technology · ShanghaiTech University +2

Policy Optimization Continuous Control Online Mirror Descent Schrödinger Bridges RL for Diffusion Models

Mar 20, 2026

PolicySim: An LLM-Based Agent Social Simulation Sandbox for Proactive Policy Optimization

Social platforms serve as central hubs for information exchange, where user behaviors and platform interventions jointly shape opinions. However, intervention policies like recommendation and content filtering, can unintentionally amplify echo chambers and polarization, posing significant societal risks. Proactively evaluating the impact of such policies is therefore crucial. Existing approaches primarily rely on reactive online A/B testing, where risks are identified only after deployment, making risk identification delayed and costly. LLM-based social simulations offer a promising pre-deployment alternative, but current methods fall short in realistically modeling platform interventions and incorporating feedback from the platform. Bridging these gaps is essential for building actionable frameworks to assess and optimize platform policies. To this end, we propose PolicySim, an LLM-based social simulation sandbox for the proactive assessment and optimization of intervention policies. PolicySim models the bidirectional dynamics between user behavior and platform interventions through two key components: (1) a user agent module refined via supervised fine-tuning (SFT) and direct preference optimization (DPO) to achieve platform-specific behavioral realism; and (2) an adaptive intervention module that employs a contextual bandit with message passing to capture dynamic network structures. Experiments show that PolicySim can accurately simulate platform ecosystems at both micro and macro levels and support effective intervention policy.

Renhong Huang, Ning Tang, Jiarong Xu +6

Zhejiang University, Hangzhou, China · Fudan University, Shanghai, China · HKUST, Hong Kong, China +3

Large Language Model-Based Social Simulation Policy Optimization User Simulation Social Media Analysis

Mar 19, 2026

Adaptive Layerwise Perturbation: Unifying Off-Policy Corrections for LLM RL

Off-policy problems such as policy staleness and training--inference mismatch have become a major bottleneck for training stability and further exploration in LLM RL. The distribution gap between the inference and updated policies grows because of the techniques to enhance inference efficiency, leading to heavy-tailed importance ratios. Heavy-tailed ratios arise when the policy is locally sharp, which further inflates gradients and can push updates outside the trust region. To address this, we propose Adaptive Layerwise Perturbation (ALP), which injects small learnable perturbations into the input hidden states of each layer during updates and uses the resulting perturbed policy as the numerator of the importance ratio against the unchanged inference policy in the objective. Intuitively, by adding controlled noise to intermediate representations, ALP prevents the updated policy from deviating too sharply from the inference policy and enlarges the policy family to cover inference-time mismatch noise. Hence, the flattened distribution can naturally tighten the gap between the updated and inference policies and reduce the tail of importance ratios, thus maintaining training stability. This is further validated empirically. Experiments on single-turn math and multi-turn tool-integrated reasoning tasks show that ALP not only improves final performance, but also avoids blow-up in the importance-ratio tail and KL spikes during iterative training, along with boosted exploration. Ablations show that representation-level perturbations across all layers are most effective, substantially outperforming partial-layer and logits-only variants.

Chenlu Ye, Xuanchang Zhang, Yifan Hao +6

University of Illinois Urbana-Champaign · Amazon

Importance Sampling for RL RL for Language Models Reinforcement Learning Off-Policy RL RL for Language Model Reasoning Policy Optimization

Mar 19, 2026

Evaluating Model-Free Policy Optimization in Masked-Action Environments via an Exact Blackjack Oracle

Infinite-shoe casino blackjack provides a rigorous, exactly verifiable benchmark for discrete stochastic control under dynamically masked actions. Under a fixed Vegas-style ruleset (S17, 3:2 payout, dealer peek, double on any two, double after split, resplit to four), an exact dynamic programming (DP) oracle was derived over 4,600 canonical decision cells. This oracle yielded ground-truth action values, optimal policy labels, and a theoretical expected value (EV) of -0.00161 per hand. To evaluate sample-efficient policy recovery, three model-free optimizers were trained via simulated interaction: masked REINFORCE with a per-cell exponential moving average baseline, simultaneous perturbation stochastic approximation (SPSA), and the cross-entropy method (CEM). REINFORCE was the most sample-efficient, achieving a 46.37% action-match rate and an EV of -0.04688 after 10^6 hands, outperforming CEM (39.46%, 7.5x10^6 evaluations) and SPSA (38.63%, 4.8x10^6 evaluations). However, all methods exhibited substantial cell-conditional regret, indicating persistent policy-level errors despite smooth reward convergence. This gap shows that tabular environments with severe state-visitation sparsity and dynamic action masking remain challenging, while aggregate reward curves can obscure critical local failures. As a negative control, it was proven and empirically confirmed that under i.i.d. draws without counting, optimal bet sizing collapses to the table minimum. In addition, larger wagers strictly increased volatility and ruin without improving expectation. These results highlight the need for exact oracles and negative controls to avoid mistaking stochastic variability for genuine algorithmic performance.

Kevin Song

Department of Biomedical Engineering The University of Alabama at Birmingham

Policy Optimization Reinforcement Learning Model-Free RL Sample-Efficient RL

Mar 18, 2026

Flow Matching Policy Optimization with Mirror Descent and Entropy Constraints

Balancing policy expressiveness with the exploration-exploitation trade-off is a core challenge in online Reinforcement Learning (RL). While Stochastic Differential Equation (SDE)-based diffusion policies can represent complex, multimodal action distributions, they suffer from two critical limitations: their stochastic reverse processes render entropy intractable (necessitating heuristic exploration), and computing policy gradients through long denoising chains is expensive and unstable. In this work, we show that ODE-based flow matching inherently resolves these issues by enabling both simulation-free policy optimization and tractable entropy computation. Building on this, we introduce Flow Matching Policy Optimization with Mirror Descent and Entropy Constraints (FMER). Our framework exploits this insight in three ways. First, we derive an analytic entropy objective that corrects for the density distortion caused by the

\tanh

transformation (mapping an unbounded latent space to bounded actions), thereby facilitating entropy-constrained policy optimization. Second, we theoretically establish that minimizing an advantage-weighted conditional flow matching loss acts as a simulation-free surrogate for policy mirror descent. This steers the velocity field toward high-value regions while entirely avoiding backpropagation through the ODE solver. Finally, we dynamically tune the mirror descent temperature based on the effective sample size to enforce a robust trust region during training. Empirical evaluations demonstrate that FMER achieves superior performance on the challenging sparse-reward FrankaKitchen environment, while maintaining competitive results across standard dense-reward MuJoCo benchmarks. Code is available at https://github.com/gtanaelle/FMER.git.

Ting Gao, Stavros Orfanoudakis, Nan Lin +3

Policy Optimization Maximum Entropy RL

Mar 18, 2026

AR-CoPO: Align Autoregressive Video Generation with Contrastive Policy Optimization

Streaming autoregressive (AR) video generators combined with few-step distillation achieve low-latency, high-quality synthesis, yet remain difficult to align via reinforcement learning from human feedback (RLHF). Existing SDE-based GRPO methods face challenges in this setting: few-step ODEs and consistency model samplers deviate from standard flow-matching ODEs, and their short, low-stochasticity trajectories are highly sensitive to initialization noise, rendering intermediate SDE exploration ineffective. We propose AR-CoPO (AutoRegressive Contrastive Policy Optimization), a framework that adapts the Neighbor GRPO contrastive perspective to streaming AR generation. AR-CoPO introduces chunk-level alignment via a forking mechanism that constructs neighborhood candidates at a randomly selected chunk, assigns sequence-level rewards, and performs localized GRPO updates. We further propose a semi-on-policy training strategy that complements on-policy exploration with exploitation over a replay buffer of reference rollouts, improving generation quality across domains. Experiments on Self-Forcing demonstrate that AR-CoPO improves both out-of-domain generalization and in-domain human preference alignment over the baseline, providing evidence of genuine alignment rather than reward hacking.

Dailan He, Guanlin Feng, Xingtong Ge +5

CUHK MMLab, China · Vivix Group Limited, China · HKUST, China +1

RL for Video Generation Streaming Video Generation Policy Optimization RL for Generative Models Autoregressive Video Generation

Mar 13, 2026

Optimize Wider, Not Deeper: Consensus Aggregation for Policy Optimization

Proximal policy optimization (PPO) approximates the trust region update using multiple epochs of clipped SGD. Each epoch may drift further from the natural gradient direction, creating path-dependent noise. To understand this drift, we can use Fisher information geometry to decompose policy updates into signal (the natural gradient projection) and waste (the Fisher-orthogonal residual that consumes trust region budget without first-order surrogate improvement). Empirically, signal saturates but waste grows with additional epochs, creating an optimization-depth dilemma. We propose Consensus Aggregation for Policy Optimization (CAPO), which redirects compute from depth to width:

K

PPO replicates are optimized on the same batch, differing only in minibatch shuffling order, and then aggregated into a consensus. We study aggregation in two spaces: Euclidean parameter space, and the natural parameter space of the policy distribution via the logarithmic opinion pool. In natural parameter space, the consensus provably achieves higher KL-penalized surrogate and tighter trust region compliance than the mean expert; parameter averaging inherits these guarantees approximately. On continuous control tasks, CAPO outperforms PPO and compute-matched deeper baselines under fixed sample budgets by up to 8.6x. CAPO demonstrates that policy optimization can be improved by optimizing wider, rather than deeper, without additional environment interactions.

Zelal Su, Mustafaoglu, Sungyoung Lee +3

Lain · Department of Computer Science, University of Texas at Austin · Department of Electrical and Computer Engineering, University of Texas at Austin

Policy Optimization Proximal Policy Optimization Trust Region Policy Optimization

Mar 10, 2026

Actor-Accelerated Policy Dual Averaging for Reinforcement Learning in Continuous Action Spaces

Policy Dual Averaging (PDA) offers a principled Policy Mirror Descent (PMD) framework that more naturally admits value function approximation than standard PMD, enabling the use of approximate advantage (or Q-) functions while retaining strong convergence guarantees. However, applying PDA in continuous state and action spaces remains computationally challenging, since action selection involves solving an optimization sub-problem at each decision step. In this paper, we propose actor-accelerated PDA, which uses a learned policy network to approximate the solution of the optimization sub-problems, yielding faster runtimes while maintaining convergence guarantees. We provide a theoretical analysis that quantifies how actor approximation error impacts the convergence of PDA under suitable assumptions. We then evaluate its performance on several benchmarks in robotics, control, and operations research problems. Actor-accelerated PDA achieves superior performance compared to popular on-policy baselines such as Proximal Policy Optimization (PPO). Overall, our results bridge the gap between the theoretical advantages of PDA and its practical deployment in continuous-action problems with function approximation.

Ji Gao, Caleb Ju, Guanghui Lan +1

School of Chemical and Biomolecular Engineering, Georgia Institute of Technology, GA, USA. · H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology, GA, USA. · H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology, GA, USA

Reinforcement Learning Policy Optimization Policy Learning

Mar 9, 2026

Fibration Policy Optimization

Large language models are increasingly trained as heterogeneous systems spanning multiple domains, expert partitions, and agentic pipelines, yet prevalent proximal objectives operate at a single scale and lack a principled mechanism for coupling token-level, trajectory-level, and higher-level hierarchical stability control. To bridge this gap, we derive the Aggregational Policy Censoring Objective (APC-Obj), the first exact unconstrained reformulation of sample-based TV-TRPO, establishing that clipping-based surrogate design and trust-region optimization are dual formulations of the same problem. Building on this foundation, we develop Fiber Bundle Gating (FBG), an algebraic framework that organizes sampled RL data as a fiber bundle and decomposes ratio gating into a base-level gate on trajectory aggregates and a fiber-level gate on per-token residuals, with provable first-order agreement with the true RL objective near on-policy. From APC-Obj and FBG we derive Fibration Policy Optimization (or simply, FiberPO), a concrete objective whose Jacobian is block-diagonal over trajectories, reduces to identity at on-policy, and provides better update direction thus improving token efficiency. The compositional nature of the framework extends beyond the trajectory-token case: fibrations compose algebraically into a Fibration Gating Hierarchy (FGH) that scales the same gating mechanism to arbitrary hierarchical depth without new primitives, as demonstrated by FiberPO-Domain, a four-level instantiation with independent trust-region budgets at the domain, prompt group, trajectory, and token levels. Together, these results connect the trust-region theory, a compositional algebraic structure, and practical multi-scale stability control into a unified framework for LLM policy optimization.

Chang Li, Tshihao Tsu, Yaren Zhang +2

JD Explore Academy · Carleton University

Policy Optimization RL for Language Models RL Fine-Tuning Trust Region Policy Optimization Hierarchical RL

Mar 6, 2026

Reference-guided Policy Optimization for Molecular Optimization via LLM Reasoning

Large language models (LLMs) benefit substantially from supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR) in reasoning tasks. However, these recipes perform poorly in instruction-based molecular optimization, where each data point typically provides only a single optimized reference molecule and no step-by-step optimization trajectory. We reveal that answer-only SFT on the reference molecules collapses reasoning, and RLVR provides sparse feedback under similarity constraints due to the model's lack of effective exploration, which slows learning and limits optimization. To encourage the exploration of new molecules while balancing the exploitation of the reference molecules, we introduce Reference-guided Policy Optimization (RePO), an optimization approach that learns from reference molecules without requiring trajectory data. At each update, RePO samples candidate molecules with their intermediate reasoning trajectories from the model and trains the model using verifiable rewards that measure property satisfaction under similarity constraints in an RL manner. Meanwhile, it applies reference guidance by keeping the policy's intermediate reasoning trajectory as context and training only the answer in a supervised manner. Together, the RL term promotes exploration, while the guidance term mitigates reward sparsity and stabilizes training by grounding outputs to references when many valid molecular edits exist. Across molecular optimization benchmarks, RePO consistently outperforms SFT and RLVR baselines (e.g., GRPO), achieving improvements on the optimization metric (Success Rate

\times

Similarity), improving balance across competing objectives, and generalizing better to unseen instruction styles. Our code is publicly available at https://github.com/tmlr-group/RePO.

Xuan Li, Zhanke Zhou, Zongze Li +4

Hong Kong Baptist University · CMIC, Shanghai Jiao Tong University · DAMO Academy, Alibaba Group +1

Molecular Optimization Large Language Model-Guided Optimization Policy Optimization RL for Language Models Reinforcement Learning with Verifiable Rewards

Mar 5, 2026

EvoTool: Self-Evolving Tool-Use Policy Optimization in LLM Agents via Blame-Aware Mutation and Diversity-Aware Selection

LLM-based agents depend on effective tool-use policies to solve complex tasks, yet optimizing these policies remains challenging due to delayed supervision and the difficulty of credit assignment in long-horizon trajectories. Existing optimization approaches tend to be either monolithic, which are prone to entangling behaviors, or single-aspect, which ignore cross-module error propagation. To address these limitations, we propose EvoTool, a self-evolving framework that optimizes a modular tool-use policy via a gradient-free evolutionary paradigm. EvoTool decomposes agent's tool-use policy into four modules, including Planner, Selector, Caller, and Synthesizer, and iteratively improves them in a self-improving loop through three novel mechanisms. Trajectory-Grounded Blame Attribution uses diagnostic traces to localize failures to a specific module. Feedback-Guided Targeted Mutation then edits only that module via natural-language critique. Diversity-Aware Population Selection preserves complementary candidates to ensure solution diversity. Across four benchmarks, EvoTool outperforms strong baselines by over 5 points on both GPT-4.1 and Qwen3-8B, while achieving superior efficiency and transferability. The code will be released once paper is accepted.

Shuo Yang, Soyeon Caren Han, Xueqi Ma +3

School of Computing and Information Systems, The University of Melbourne

Policy Optimization LLM Agent Self-Improvement Self-Evolving Agents Evolutionary Optimization RL for Tool Use

Mar 4, 2026

HALO: Learning Human-Robot Collaboration via Heterogeneous-Agent Lyapunov Policy Optimization

To improve generalization and resilience in human-robot collaboration (HRC), robots must contend with diverse combinations of human behaviors and contexts, motivating multi-agent reinforcement learning (MARL). However, inherent heterogeneity between robots and humans creates a rationality gap (RG), where decentralized policy updates deviate from cooperative joint optimization. The resulting learning problem is a general-sum differentiable game, so independent policy-gradient updates can oscillate or diverge without added structure. We propose heterogeneous-agent Lyapunov policy optimization (HALO), a framework that stabilizes decentralized MARL by enforcing Lyapunov-based contraction in policy-parameter space. Unlike Lyapunov-based safe RL, which targets state/trajectory constraints in constrained Markov decision processes, HALO uses Lyapunov certification to stabilize decentralized policy learning. HALO rectifies decentralized gradients via optimal quadratic projections, ensuring monotonic contraction of RG and enabling effective exploration of open-ended interaction spaces. Extensive simulations and real-world humanoid-robot experiments show that this certified stability improves generalization and robustness in collaborative corner cases. Our project website is available at https://HaoZhang-THU.github.io/HALO/.

Hao Zhang, Yaru Niu, Yikai Wang +2

ETAIC Lab, Department of Electrical Engineering, University of Texas at Arlington, USA · Safe AI Lab, Department of Mechanical Engineering, Carnegie Mellon University, USA.

Decentralized MARL Heterogeneous MARL Policy Optimization Human-Robot Collaboration Robot Policy Generalization

Mar 3, 2026

Wasserstein Proximal Policy Gradient

We study policy gradient methods for continuous-action, entropy-regularized reinforcement learning through the lens of Wasserstein geometry. Starting from a Wasserstein proximal update, we derive Wasserstein Proximal Policy Gradient (WPPG) via an operator-splitting scheme that alternates an optimal transport update with a heat step implemented by Gaussian convolution. This formulation avoids evaluating the policy's log density or its gradient, making the method directly applicable to expressive implicit stochastic policies specified as pushforward maps. We establish a global linear convergence rate for WPPG, covering both exact policy evaluation and actor-critic implementations with controlled approximation error. Empirically, WPPG is simple to implement and attains competitive performance on standard continuous-control benchmarks.

Zhaoyu Zhu, Shuhan Zhang, Rui Gao +1

Zhiyuan College, Shanghai Jiao Tong University, Shanghai 200240, China · School of Data Science, The Chinese University of Hong Kong, Shenzhen, Guangdong, China · McCombs School of Business, The University of Texas at Austin, Austin, TX, USA

Policy Optimization Policy Gradient Methods Maximum Entropy RL Wasserstein Gradient Flows Actor-Critic Methods

Mar 2, 2026

Efficient Token-Level Policy Optimization for Masked Diffusion Models via Velocity Alignment

Reinforcement Learning with Verifiable Rewards (RLVR) successfully aligns autoregressive models in rigorous domains like mathematical reasoning and code generation. However, applying RLVR to Diffusion Large Language Models (dLLMs) is hindered by intractable sequence-level likelihoods, forcing existing methods to rely on unstable, high-variance approximations. We propose Sequence Likelihood-Free Policy Optimization (LFPO), an efficient framework that circumvents this bottleneck by formulating policy alignment as discrete velocity alignment. Inspired by continuous flow matching, LFPO eschews trajectory-level estimation and directly optimizes denoising logits via contrastive token-level updates, providing a stable, low-variance surrogate. Furthermore, predicting terminal solutions from intermediate states implicitly straightens the generation trajectory, enabling high-quality sampling with significantly fewer denoising iterations. Extensive experiments on reasoning and coding benchmarks demonstrate that LFPO matches or exceeds the accuracy of likelihood-based baselines while substantially accelerating inference.

Chenxing Wei, Jiazhen Kang, Hong Wang +8

Shenzhen University, China · Guangming Laboratory, China · ByteDance, China +5

RL for Diffusion Models Policy Optimization Diffusion Language Model Inference Masked Diffusion Models Reinforcement Learning with Verifiable Rewards Diffusion Model Inference Acceleration

Mar 1, 2026

Stabilizing Policy Optimization via Logits Convexity

While reinforcement learning (RL) has been central to the recent success of large language models (LLMs), RL optimization is notoriously unstable, especially when compared to supervised fine-tuning (SFT). In this work, we investigate the stability gap between SFT and RL from a gradient-based perspective, and show that the convexity of the SFT loss with respect to model logits plays a key role in enabling stable training. Our theoretical analysis demonstrates that this property induces favorable gradient directionality during optimization. In contrast, Proximal Policy Optimization (PPO), a widely adopted policy gradient algorithm utilizing a clipped surrogate objective, lacks this stabilizing property. Motivated by this observation, we propose Logits Convex Optimization (LCO), a simple yet effective policy optimization framework that aligns the learned policy with an optimal target derived from the original RL objective, thereby emulating the stabilizing effects of logits-level convexity. Extensive experiments across multiple model families show that our LCO framework consistently improves training stability and outperforms conventional RL methods on a broad range of benchmarks.

Hongzhan Chen, Tao Yang, Yuhua Zhu +3

School of Computer Science and Engineering, Sun Yat-sen University, China · Shanghai Innovation Institute · Wechat Search, Tencent Inc, China +1

Policy Optimization Reinforcement Learning RL Fine-Tuning Policy Gradient Methods Proximal Policy Optimization

Feb 28, 2026

MemPO: Self-Memory Policy Optimization for Long-Horizon Agents

Long-horizon agents face the challenge of growing context size during interaction with environment, which degrades the performance and stability. Existing methods typically introduce the external memory module and look up the relevant information from the stored memory, which prevents the model itself from proactively managing its memory content and aligning with the agent's overarching task objectives. To address these limitations, we propose the self-memory policy optimization algorithm (MemPO), which enables the agent (policy model) to autonomously summarize and manage their memory during interaction with environment. By improving the credit assignment mechanism based on memory effectiveness, the policy model can selectively retain crucial information, significantly reducing token consumption while preserving task performance. Extensive experiments and analyses confirm that MemPO achieves absolute F1 score gains of 25.98 over the base model and 7.1 over the previous SOTA baseline, while reducing token usage by 67.58% and 73.12%. The code is released at https://github.com/TheNewBeeKing/MemPO.

Ruoran Li, Xinghua Zhang, Haiyang Yu +7

Tsinghua University · Tongyi Lab, Alibaba Group

Agent Memory Management Memory-Augmented Agents Long-Horizon Agent Tasks Policy Optimization RL for LLM Agents

Feb 22, 2026

Smooth Gate Functions for Soft Advantage Policy Optimization

Group Relative Policy Optimization (GRPO) has significantly advanced the training of large language models and enhanced their reasoning capabilities, while it remains susceptible to instability due to the use of hard clipping. Soft Adaptive Policy Optimization (SAPO) addresses this limitation by replacing clipping with a smooth sigmoid-based gate function, which leads to more stable updates. We have decided to push this theory further and investigate the impact of different gate functions on both training stability and final model performance. We formalize the key properties that admissible gates should satisfy and identify several families of such functions for empirical evaluation. This paper presents an analysis of our findings based on experiments conducted with the Qwen2.5-7B-Instruct model on mathematical reasoning tasks. These results provide practical guidance for designing smoother and more robust policy optimization objectives for large language model training.

Egor Denisov, Svetlana Glazyrina, Maksim Kryzhanovskiy +1

Lomonosov Moscow State University, Moscow, Russia · Institute for Artificial Intelligence, Lomonosov Moscow State University, Moscow, Russia

Policy Optimization RL for Language Model Reasoning Group Relative Policy Optimization

Feb 19, 2026

MePoly: Max Entropy Polynomial Policy Optimization

Stochastic Optimal Control provides a unified mathematical framework for solving complex decision-making problems, encompassing paradigms such as maximum entropy reinforcement learning(RL) and imitation learning(IL). However, conventional parametric policies often struggle to represent the multi-modality of the solutions. Though diffusion-based policies are aimed at recovering the multi-modality, they lack an explicit probability density, which complicates policy-gradient optimization. To bridge this gap, we propose MePoly, a novel policy parameterization based on polynomial energy-based models. MePoly provides an explicit, tractable probability density, enabling exact entropy maximization. Theoretically, we ground our method in the classical moment problem, leveraging the universal approximation capabilities for arbitrary distributions. Empirically, we demonstrate that MePoly effectively captures complex non-convex manifolds and outperforms baselines in performance across diverse benchmarks.

Hang Liu, Sangli Teng, Maani Ghaffari

University of Michigan, Ann Arbor · University of California, Berkeley

Policy Optimization Maximum Entropy RL

Feb 16, 2026

Beyond Token-Level Policy Gradients for Complex Reasoning with Large Language Models

Existing policy-gradient methods for auto-regressive language models typically select subsequent tokens one at a time as actions in the policy. While effective for many generation tasks, such an approach may not fully capture the structure of complex reasoning tasks, where a single semantic decision is often realized across multiple tokens--for example, when defining variables or composing equations. This introduces a potential mismatch between token-level optimization and the inherently block-level nature of reasoning in these settings. To bridge this gap, we propose Multi-token Policy Gradient Optimization (MPO), a framework that treats sequences of K consecutive tokens as unified semantic actions. This block-level perspective enables our method to capture the compositional structure of reasoning trajectories and supports optimization over coherent, higher-level objectives. Experiments on mathematical reasoning and coding benchmarks show that MPO outperforms standard token-level policy gradient baselines, highlight the limitations of token-level policy gradients for complex reasoning, motivating future research to look beyond token-level granularity for reasoning-intensive language tasks.

Mufan Xu, Kehai Chen, Xuefeng Bai +4

School of Computer Science and Technology, Harbin Institute of Technology, China · Baidu Inc., Beijing, China

RL for Language Model Reasoning Policy Optimization LLM Reasoning

Feb 6, 2026

Difficulty-Estimated Policy Optimization

Recent advancements in Large Reasoning Models (LRMs), exemplified by DeepSeek-R1, have underscored the potential of scaling inference-time compute through Group Relative Policy Optimization (GRPO). However, GRPO frequently suffers from gradient signal attenuation when encountering problems that are either too trivial or overly complex. In these scenarios, the disappearance of inter-group advantages makes the gradient signal susceptible to noise, thereby jeopardizing convergence stability. While variants like DAPO attempt to rectify gradient vanishing, they do not alleviate the substantial computational overhead incurred by exhaustive rollouts on low-utility samples. In this paper, we propose Difficulty-Estimated Policy Optimization (DEPO), a novel framework designed to optimize the efficiency and robustness of reasoning alignment. DEPO integrates an online Difficulty Estimator that dynamically assesses and filters training data before the rollout phase. This mechanism ensures that computational resources are prioritized for samples with high learning potential. Empirical results show that DEPO achieves approximately 1.86x faster rollouts than DAPO in the profiled configuration, while improving benchmark performance over GRPO at comparable training cost. Our approach significantly lowers the computational barrier for training high-performance reasoning models, offering a more sustainable path for reasoning scaling.

Yu Zhao, Fan Jiang, Tianle Liu +5

Alibaba International Digital Commerce · School of Software Technology, Dalian University of Technology Dalian, China

RL for Language Model Reasoning Policy Optimization Reinforcement Learning Group Relative Policy Optimization

Feb 4, 2026

QUATRO: Query-Adaptive Trust Region Policy Optimization for LLM Fine-tuning

GRPO-style reinforcement learning (RL)-based LLM fine-tuning algorithms have recently gained popularity. Relying on heuristic trust-region approximations, however, they can lead to brittle optimization behavior, as global importance-ratio clipping and group-wise normalization fail to regulate samples whose importance ratios fall outside the clipping range. We propose Query-Adaptive Trust-Region policy Optimization (QUATRO), which directly enforces trust-region constraints through a principled optimization. This yields a clear and interpretable objective that enables explicit control over policy updates and stable, entropy-controlled optimization, with a stabilizer terms arising intrinsically from the exact trust-region formulation. Empirically verified on diverse mathematical reasoning benchmarks, QUATRO shows stable training under increased policy staleness and aggressive learning rates, maintaining well-controlled entropy throughout training.

Doyeon Lee, Eunyi Lyou, Hyunsoo Cho +3

Seoul National University · Department of Artificial Intelligence / Institute for Multiscale Matter and Systems, Ewha Woman University · Geogia Institute of Technology

Trust-Region Optimization Policy Learning Policy Optimization RL for Language Model Reasoning Trust Region Policy Optimization RL Fine-Tuning+1

Feb 2, 2026

Data- and Variance-dependent Regret Bounds for Online Tabular MDPs

This work studies online episodic tabular Markov decision processes (MDPs) with known transitions and develops best-of-both-worlds algorithms that achieve refined data-dependent regret bounds in the adversarial regime and variance-dependent regret bounds in the stochastic regime. We quantify MDP complexity using a first-order quantity and several new data-dependent measures for the adversarial regime, including a second-order quantity and a path-length measure, as well as variance-based measures for the stochastic regime. To adapt to these measures, we develop algorithms based on global optimization and policy optimization, both built on optimistic follow-the-regularized-leader with log-barrier regularization. For global optimization, our algorithms achieve first-order, second-order, and path-length regret bounds in the adversarial regime, and in the stochastic regime, they achieve a variance-aware gap-independent bound and a variance-aware gap-dependent bound that is polylogarithmic in the number of episodes. For policy optimization, our algorithms achieve the same data- and variance-dependent adaptivity, up to a factor of the episode horizon, by exploiting a new optimistic

Q

-function estimator. Finally, we establish regret lower bounds in terms of data-dependent complexity measures for the adversarial regime and a variance measure for the stochastic regime, implying that the regret upper bounds achieved by the global-optimization approach are nearly optimal.

Mingyi Li, Taira Tsuchiya, Kenji Yamanishi

Department of Mathematical Informatics, The University of Tokyo, Tokyo, Japan · RIKEN AIP, Tokyo, Japan

Regret Minimization in RL Markov Decision Processes Reinforcement Learning Policy Optimization Follow-the-Regularized-Leader

Jan 30, 2026

RN-D: Discretized Categorical Actors for On-Policy Reinforcement Learning

On-policy Reinforcement Learning (RL) remains a dominant paradigm for continuous control, yet standard implementations rely on Gaussian actors and relatively shallow MLP policies, often leading to brittle optimization when gradients are noisy, and policy updates must be conservative. In this paper, we revisit actor policy representation as a first-class design choice for on-policy RL. We study discretized categorical actors, which represent each action dimension as a distribution over discrete bins and induce a policy objective analogous to classification cross-entropy loss. Building on architectural advances from supervised learning, we further pair discretized categorical actors with regularized networks, yielding RN-D. Across diverse continuous-control benchmarks, we show that simply replacing the standard Gaussian actor with our proposed actor substantially improves performance, achieving state-of-the-art results within on-policy RL. We release our code at https://github.com/alwaysbyx/RND-RL.

Yuexin Bian, Jie Feng, Tao Wang +3

University of California San Diego, La Jolla, USA

Policy Optimization Continuous Control

Jan 30, 2026

Cantelli Constrained Policy Optimization

We introduce Canary, a risk-averse method designed to optimize Value-at-Risk (VaR) constrained reinforcement learning (RL) problems. We employ Cantelli's inequality to obtain a tractable, conservative and smooth bound on the VaR constraint based on the first two moments of the cost return. This yields a constraint estimator that remains stable with tight violation thresholds in dense cost regimes. Extending the trust-region framework of the Constrained Policy Optimization (CPO) method, we further provide worst-case bounds for both policy improvement and constraint violation during the training process. Empirically during training, Canary is the only method that reliably satisfies the VaR constraint in every environment tested.

Rohan Tangri, Jan-Peter Calliess

Machine Learning Research Group, University of Oxford, Oxford, United Kingdom

Policy Optimization Constrained RL Risk-Sensitive RL