Policy Optimization

Ask about this topic

Overview

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

CardsList

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

Jun 29, 2026

Predictable GRPO: A Closed-Form Model of Training Dynamics

We develop a first-principles reduced-order model of these dynamics. Under a single mean-field assumption that summarizes the policy by its expected reward, we reduce the GRPO update to a stochastically-forced damped oscillator whose mass, damping, and stiffness are fixed in closed form by the optimizer hyperparameters together with a single measured curvature scale -- momentum supplies the inertia, off-policy lag erodes the damping, and the group size enters, to leading order, as a noise temperature. The reduction has three consequences. First, it subsumes the empirical single-exponential saturation law as its overdamped limit, recasting the fitted plateau, timescale, and size exponent as the fixed point, inverse stiffness, and curvature-scaling exponent of the underlying potential, and adding, through the retained inertial term, the slow-start phase the single exponential cannot represent. Second, it yields predictions tied to independently measurable quantities rather than fitted ones: group-size invariance of the deterministic trajectory with a 1/G1/G stationary fluctuation, a sharp stability threshold in the refresh interval, and an overdamped-to-oscillatory transition. Third, it furnishes diagnostics that separate failure modes a reward curve alone conflates -- reward hacking, advantage degeneracy, policy concentration, and dynamical instability. Across three models and two group sizes, the closed-form trajectory fits training reward to R2≥0.91R^2 \geq 0.91 and the mean trajectory is group-size invariant to leading order -- on both the reward curve and out-of-distribution transfer to eight math benchmarks -- while the within-group reward spread retains a residual GG-dependence that the leading-order temperature picture does not capture.

Rajat Ghosh, Datta Nimmaturi, Aryan Singhal +4

Nutanix San Jose, USA · Unsloth Hyderabad, India · Nutanix2026 San Jose, USA +1

Group Relative Policy OptimizationPolicy OptimizationNeural Network Training DynamicsPolicy LearningGradient Descent Dynamics
AskPDF
Jun 28, 2026

Hierarchical Policy Learning via Spectral Decomposition

In this paper, we identify a semantic decomposition in robot action sequences, separating task-level motion intent from execution-level refinements. By analyzing actions in the spectral domain using the discrete cosine transform (DCT), we observe that low-frequency components capture global motion trajectories, while high-frequency components encode precise timing, alignment, and contact behaviors. Motivated by this structure, we propose Causal Spectral Policy (CSP), which models action generation as a causal coarse-to-fine process: coarse motion is predicted from observation and language, and fine corrections are generated conditionally on the realized trajectory. Across simulation and real-world evaluations, CSP consistently outperforms strong baselines on precision-sensitive manipulation tasks. Additionally, we propose human-inspired teleoperation noise injection as a data augmentation method, under which our approach demonstrates strong robustness to noisy demonstrations.

Shuxin Cao, Liquan Wang, Walker Byrnes +3

Georgia Institute of Technology, Atlanta, Georgia, USA · Georgia Tech Research Institute, Atlanta, Georgia, USA · Harvard University, Cambridge, Massachusetts, USA

Robot Policy LearningRobot Skill LearningLanguage-Conditioned Robot ManipulationHierarchical RLPolicy OptimizationRobotic Manipulation
AskPDF
Jun 28, 2026

The Mirage of Optimizing Training Policies: Monotonic Inference Policies as the Real Objective for LLM Reinforcement Learning

Reinforcement learning (RL) has gained growing attention in large language model (LLM) post-training, yet RL training remains fragile and can suffer from instability or collapse. One vital cause is training-inference mismatch: LLM adopts separate inference and training engines for generation efficiency and training precision, which in practice exhibits inconsistent probabilities for the same trajectories on training and inference sides, even with synchronized model parameters. This naturally induces a special type of off-policyness ever existing and poisoning the training. Prior works have made various efforts in addressing the off-policyness to stabilize the training policies under the mismatch. In this paper, we point out the objective misalignment neglected by existing works that an effective update to the policy in the training engine not necessarily ensures the improvement of the inference policy, i.e., the one used in deployment. To this end, we propose a new policy optimization objective for LLM RL, named Monotonic Inference Policy Improvement (MIPI). Following this principle, we introduce Monotonic Inference Policy Update (MIPU), a two-step LLM RL framework that constructs sampler-referenced candidate updates and selectively accepts synchronized candidates using an inference-side gap proxy. Experiments conducted on two model scales under high mismatch show that MIPU improves average reasoning performance and training stability.

Jing Liang, Hongyao Tang, Yi Ma +9

1Tianjin · 2Alibaba

Reinforcement LearningRL for Language Model ReasoningPolicy OptimizationRL Post-TrainingOff-Policy RL
AskPDF
Jun 24, 2026

MiniOpt: Reasoning to Model and Solve General Optimization Problems with Limited Resources

Achieving strong optimization generalization across diverse optimization problems while requiring limited training resources remains a challenging problem for optimization-oriented large language models (LLMs). Existing approaches typically rely on large-scale supervised datasets, costly reasoning annotations, and expensive intermediate step verification, resulting in substantial training overhead. To address these challenges, we propose MiniOpt, a reinforcement learning framework that learns to solve optimization problems through an "reasoning-to-model-and-solve" paradigm. MiniOpt decomposes optimization reasoning into structured optimization modeling and executable solver generation. Building upon this paradigm, we introduce OptReward, a reward function with hierarchical score structure that jointly evaluates formulation and solution, enabling effective policy learning without expert demonstrations. We further develop an optimization-oriented policy optimization strategy that improves exploration efficiency and stabilizes reinforcement learning for compact models. Extensive experiments show that MiniOpt-3B exhibits strong optimization generalization across various optimization types, problem scenarios, and task domains. For models with fewer than 10B parameters, MiniOpt series achieves the highest average solving accuracy (SA). For models with more than 10B parameters, MiniOpt still shows competitive performance. These results suggest that optimization-oriented reward design and reinforcement learning provide an effective pathway for developing compact optimization-specialized language models with strong optimization generalization capabilities. The code is available at https://github.com/Hsiang-1/MiniOpt.

Ke Zhao, Zixiang Di, Hong Qian +9

East China Normal University, Shanghai 200062, China · AntGroup, Hangzhou 310000, China · Southern University of Science and Technology, Shenzhen 518055, China +1

Large Language Model-Guided OptimizationRL for Language Model ReasoningPolicy OptimizationNatural Language to Optimization ModelingReward Design for RL
AskPDF
Jun 19, 2026

Entropy Objectives in Markov Decision Processes

We consider the problem of synthesizing control policies that enforce a concentration property on the state distributions of a stochastic system. We present a formalization of this problem in terms of synthesizing strategies for maintaining an entropy-based objective in Markov Decision Processes (MDPs). We first show that even relaxed versions of this problem are complexity-theoretically hard. We then present a sound and (conditionally) relatively complete method to verify and synthesize strategies for such entropy objectives. The main challenge is the non-linear nature of such objectives, and our approach addresses this by exploiting and combining ideas from convex duality and invariant synthesis. We also investigate the role of memory and randomization in ensuring entropy objectives. Finally, we implement our ideas to evaluate our approach empirically on a few illustrative benchmarks.

S. Akshay, Raghav Goyal, Aditya Neeraje +1

Indian Institute of Technology Bombay, Mumbai, India. · Tata Institute of Fundamental Research, Mumbai, India.

Markov Decision ProcessesMarkov ModelsPolicy OptimizationFormal Verification
AskPDF
Jun 19, 2026

Robot Self-Improvement via Human-Video Dynamics Models

A central question in robot learning is how to acquire skills from the kinds of data that humans learn from: passive observation, embodied practice, and the experience of failure. Human videos provide the first of these in abundance, and prior work has shown they can initialize useful policies. Far less clear is whether they can support the second and third: whether priors extracted from human videos can ground a robot's own attempts well enough to evaluate them, correct them, and improve from them. In this work, we show that human videos can be used to learn embodiment-agnostic action, dynamics, and value representations that transfer across robot embodiments, providing the predictive foundation required for robots to autonomously improve from their own rollouts and failures. We introduce Dynamics-Guided Action Correction (DGAC), a training-free approach that leverages these adapted models to repair failed states: each failure becomes a query for which the learned models propose and rank corrective actions, turning failures into supervision for the next policy update. Across seven real-world manipulation tasks spanning both a mobile manipulator and a static manipulator arm, our approach improves success rates from 40% to 81% across multiple policy backbones, demonstrating cross-embodiment robot self-improvement from human-video priors. These results show that human priors and robot failures can be combined to enable scalable autonomous policy improvement. Project page: https://ethz-mrl.github.io/robot-self-improvement-website/.

Hanzhi Chen, Anran Zhang, Simon Schaefer +5

1ETH Zurich · 2Technical University of Munich · 4MCML +1

Robot Skill LearningPolicy OptimizationCross-Embodiment Robot LearningRobotic ManipulationRobot Learning
AskPDF
Jun 18, 2026

VIMPO: Value-Implicit Policy Optimization for LLMs

Reinforcement learning with verifiable rewards has become a central tool for improving the reasoning ability of large language models, but current methods face a trade-off between simplicity and credit assignment. Group-relative methods such as GRPO avoid training a critic, but typically assign a trajectory-level advantage to every token. Actor-critic methods provide denser learning signals, but require a learned value function with its own training instability. We introduce VIMPO, a critic-free policy optimization method that derives a policy-implied value function from the optimality conditions of KL-regularized reinforcement learning. For autoregressive generation, the resulting value recurrence can be written in terms of policy-reference log-ratios and anchored by the terminal condition that no future reward remains at the end of a trajectory. This gives a simple value loss that incorporates outcome-level verifiable rewards without training a critic. The same derivation also yields a critic-free actor advantage, allowing VIMPO to separate reward incorporation through the value loss from policy improvement through a PPO-style actor update. On mathematical RLVR benchmarks, VIMPO improves over GRPO across MATH-500, AIME 2024, AIME 2025, and OlympiadBench, with especially larger gains on competition-style evaluations. Under noisy rewards, VIMPO retains a consistent advantage over GRPO, suggesting that policy-implied value optimization can provide finer credit assignment while preserving the practical simplicity of critic-free training.

Zhewei Kang, Aosong Feng, Sergey Levine +2

UC Berkeley · Yale University

RL for Language Model ReasoningPolicy OptimizationCritic-Free RLReinforcement Learning with Verifiable Rewards
AskPDF
Jun 18, 2026

Co-policy: Responsive Human-Robot Co-Creation for Musical Performances

Art has long stood as a pivotal expression of human creativity. Embodied artificial intelligence offers a route for generative models to participate in that creativity through physical action rather than disembodied digital content. In robotic music co-creation, it is challenging to connect semantic musical understanding with real-time and physically executable performance. We present Co-policy, a framework for human-robot musical co-creation that separates semantic intent grounding, constrained musical variation, and visuomotor execution. To ground musical semantics, Co-policy uses pre-inference semantic anchors and a fine-tuned Qwen-vl planner (F-Qwen) to transform speech, live musical seeds, and visual observations into structured co-creation plans. To support low-latency execution, Co-policy introduces a Gaussian-Mixture Visuomotor Policy (GMP), implemented as a conditional mixture-density policy that maps target notes and visual context to multimodal robot actions in a single forward pass. Unlike robotic playback systems that merely reproduce user-specified notes, Co-policy generates complementary musical responses under both musical and physical constraints. Real-robot chime experiments, ablations, and expert evaluation show improved intent alignment, execution accuracy, and response frequency over diffusion-policy and ablated baselines, supporting physically grounded action generation as a key requirement for embodied human-AI co-creation.

Xuetao Li, Wenke Huang, Mang Ye +4

School of Computer Science, Wuhan University, Wuhan, China · College of Computing and Data Science, Nanyang Technological University, Singapore · School of Automation, Wuhan University of Technology, Wuhan, China +2

Policy OptimizationRobot Motion GenerationHuman-Robot CollaborationHuman-Robot InteractionHuman-AI Co-Creation
AskPDF
Jun 15, 2026

A Gradient Perspective on RLVR Stability and Winner Advantage Policy Optimization

Reinforcement learning with verifiable rewards (RLVR) improves language-model reasoning, but GRPO-style optimization remains prone to collapse. We analyse this instability through token-level gradient dynamics, deriving a taxonomy that predicts how updates affect next-token probabilities and entropy. The taxonomy shows that stability depends jointly on the advantage sign and token distribution under the current policy. Motivated by this finding, we propose Winner Advantage Policy Optimization (WAPO), a simple online clipped policy-gradient objective that updates only on positive-advantage completions. Across mathematical reasoning and multi-hop QA benchmarks, WAPO improves training stability and matches or outperforms baselines across multiple model families. Full code can be found at https://github.com/layer6ai-labs/wapo.

Prasanth YSS, Zhichen Ren, Rasa Hosseinzadeh +6

Layer 6 AI

RL for Language Model ReasoningPolicy OptimizationReinforcement Learning with Verifiable RewardsPolicy Gradient Methods
AskPDF
Jun 14, 2026

Scalar-Stepsize Nonuniform Monte Carlo Optimistic Policy Iteration: A Certified Counterexample

Tsitsiklis proved convergence of Monte Carlo optimistic policy iteration under a uniform update structure and identified nonuniform update frequencies as a delicate obstruction. We give a certified negative answer for the natural scalar-stepsize, unnormalized asynchronous state-value recursion with fixed nonuniform state-selection probabilities. In a three-state, two-action discounted MDP, the nonuniform update frequencies induce a diagonally scaled greedy-policy mean field with a certified nonconstant attracting hybrid periodic orbit. With a bounded unbiased geometric-horizon estimator and Robbins--Monro stepsizes, the original stochastic recursion remains trapped near the cycle with positive probability and therefore fails to converge. The example pinpoints a geometric obstruction: uniform sampling gives radial residual contraction, whereas scalar nonuniform sampling anisotropically distorts the residual dynamics and can generate switched attracting cycles.

Yuanlong Chen

Markov Decision ProcessesPolicy OptimizationStochastic ApproximationPolicy Iteration
AskPDF
Jun 11, 2026

αα-fair heterogeneous agent reinforcement learning

Cooperation in multi-agent systems is typically optimized through utilitarian objectives that maximize overall efficiency but fail to account for reward distribution, often resulting in inequitable "leader-follower" dynamics. While fairness-based approaches encourage pro-social behaviors where every agent benefits from cooperation, many current algorithms - including those utilizing reward shaping - break the stationarity of Markov Games or lack rigorous theoretical guarantees. This creates a critical gap between fair objective methods and theoretically safe learning frameworks. We propose a novel framework that bridges αα-fairness with Heterogeneous-Agent Trust Region Learning (HATRL), ensuring monotonic improvement and convergence toward Nash Equilibria. Our approach leverages a fair advantage function that dynamically weights agent utilities based on their expected returns, allowing the global objective to transition from purely utilitarian efficiency to αα-fairness welfare based on the parameter αα. We introduce two practical algorithms, αα-fair HATRPO and αα-fair HAPPO, and demonstrate through experiments in sequential social dilemmas like CleanUp and CommonHarvest that they perform better than HATRL's algorithms from a utilitarian point of view while achieving socially higher outcomes.

Yao-hua Franck Xu, Tayeb Lemlouma, Jean-Marie Bonnin +1

Orange Innov Lannion, France · IRISA Université de Rennes Lannion, France · IRISA IMT Atlantique Rennes, France

Algorithmic FairnessPolicy OptimizationTrust Region Policy OptimizationCooperative MARLMulti-Agent Reinforcement LearningHeterogeneous MARL
AskPDF
Jun 10, 2026

IntElicit: Eliciting and Assessing Contextualized Creativity via Dialogue Policy Optimization

Contextualized assessment offers high ecological validity for evaluating creativity but introduces a critical challenge: observed performance may be confounded with cognitive proficiency (domain knowledge) and agency (willingness to engage). Meanwhile, in the age of generative AI, creative problem solving increasingly occurs in tool-mediated and human--AI interactive environments, making fully static assessment less aligned with contemporary creative practice. To address these issues, this paper proposes IntElicit, a framework for eliciting and assessing contextualized creativity via dialogue policy optimization. IntElicit functions as a constrained adaptive AI Interviewer: it provides non-directive knowledge and agency scaffolds in multi-turn interaction to reduce non-creative confounders, while preserving participants' responsibility for generating the creative content being evaluated. Specifically, to tackle sparse rewards and potential reward hacking (e.g., answer dictation) in open-ended educational dialogue, IntElicit introduces a decomposed process reward mechanism. This mechanism aligns the policy with pedagogical elicitation, rewarding prompts that draw out participant reasoning rather than producing optimal answers on their behalf. Extensive experiments, including participant simulation and a human subject study (N=64), show that IntElicit improves elicited creative outcomes over expert-designed baselines. Together, the results suggest that interactive elicitation can reveal creative potential that static FPSP-style assessment may miss, providing a formative and diagnostic lens for contextualized creativity assessment in AI-mediated learning contexts.

Mingjia Li, Jin Wu, Hong Qian +7

East China Normal University · Shanghai Innovation Institute

Creativity AssessmentPolicy OptimizationHuman-in-the-Loop EvaluationGenerative AI in EducationComputational Creativity
AskPDF
Jun 9, 2026

Test-Time Gradient Guidance of Flow Policies in Reinforcement Learning

Expressive continuous control policies, such as diffusion and flow models, form the backbone of recent advances in scaling imitation learning for simulated and real robot control. While they are known to scale stably in the supervised imitation learning setting, incorporating them into reinforcement learning (RL) pipelines for policy improvement has proven more difficult. It often requires specialized training objectives or backpropagating through denoising processes, which cause well-known issues with stability and affect scalability. In this paper we study the question of whether simple policy improvement schemes at test time alone, leaving stable supervised policy training intact, can be a competitive alternative which sidesteps these issues. To this end, we propose QGF (Q-Guided Flow), an RL algorithm that performs policy optimization entirely at test time. QGF works by pre-training both a reference flow policy (via a standard behavioral cloning objective) and a value function critic and, at test time, using the value gradient to guide the reference policy to generate higher-value actions without any additional policy learning. Empirically, QGF outperforms prior test-time RL methods on single-task and goal-conditioned offline RL benchmarks with high-dimensional action spaces, and is competitive with state-of-the-art training-time algorithms while being much cheaper to run. Moreover, it exhibits favorable scaling with model size by avoiding the instability of actor-critic training, offering a practical and effective alternative RL algorithm with expressive policies.

Zhiyuan Zhou, Andy Peng, Charles Xu +4

UC Berkeley · Physical Intelligence

Policy OptimizationOffline RLRL for Generative Models
AskPDF
Jun 9, 2026

On-sky demonstration of reinforcement learning for adaptive optics control

Reinforcement learning (RL)-based algorithms have recently emerged as a promising approach for adaptive optics (AO) control. In simulations and laboratory experiments, they have demonstrated robustness to real-world effects such as photon and detector noise, misregistration, vibrations, and rapid variations in seeing conditions. However, their performance has not yet been validated on sky. We report the first on-sky demonstration of a reinforcement learning controller for adaptive optics, named Policy Optimization for AO (PO4AO). We further analyze its on-sky behavior and identify directions for improving the algorithm and its implementation.PO4AO was implemented and deployed on the Papyrus adaptive optics system installed at the Coudé focus of the 1.52 m telescope (T152) at the OHP. A Python-based implementation was interfaced with the existing real-time controller (DAO RTC) via shared-memory buffers. The performance of PO4AO was compared to that of a standard integrator controller over several nights, covering a range of flux levels and atmospheric conditions. PO4AO consistently outperformed the standard integrator in all tested configurations. The controller successfully learned and compensated for vibration patterns and demonstrated strong robustness to measurement noise. Once tuned for Papyrus, PO4AO operated in a turnkey fashion, using a single set of hyperparameters across varying observing conditions and science targets. These performance gains were achieved despite a non-optimized Python implementation introducing approximately 750 μs750\,μ\text{s} of additional latency, along with control jitter and occasional frame drops. When properly implemented and optimized, PO4AO constitutes a robust and high-performance turnkey controller for single-conjugate adaptive optics systems, paving the way for broader adoption of reinforcement learning strategies in on-sky AO operations.

Jalo Nousiainen, Vincent Chambouleyron, Benoit Neichel +7

European Southern Observatory · Aix Marseille University, CNRS, CNES, LAM, Marseille, France · Wakea Consulting, 4 Rue Joseph Bouchayer, 38100 Grenoble +1

RL ControlReinforcement LearningPolicy OptimizationAdaptive Control
AskPDF
Jun 9, 2026

Convergence of Monte Carlo Optimistic Policy Iteration: Beyond Uniform State-Action Updates

The asymptotic behaviour of Monte Carlo optimistic policy iteration (MC-O-PI) is a long-standing open question. When the model of the environment is unknown, as is common in practice, the only known condition that guarantees convergence to optimality is impractical. In its canonical form, this condition requires that the episodes used for policy evaluation be initialised uniformly over the entire state-action space. This paper strictly relaxes that requirement. Specifically, we prove that initial-visit MC-O-PI converges to optimality even when updates are uniform only over the actions within each state. This allows episodes to start in different states at arbitrary frequencies; a realistic implementation when the state space is large or unknown but the action space in each state is manageable. The proof departs from the classical analysis of Tsitsiklis whose central commutativity argument no longer applies when states are updated at different frequencies. Instead, we first show that the mean-field dynamics of MC-O-PI generate monotonically improving policies when updates are uniform over the actions in each state, and then prove that noise cannot consistently prevent this improvement by extending the lock-in argument of the combined stability-ODE method. This approach suggests a new way to study optimistic policy-iteration algorithms in general.

Octave Oliviers, Glenn Vinnicombe

Department of Engineering University of Cambridge Cambridge, UK

Markov Decision ProcessesReinforcement LearningPolicy OptimizationPolicy IterationPolicy Learning
AskPDF
Jun 8, 2026

Rethinking the Divergence Regularization in LLM RL

Reinforcement learning (RL) has become a key component of post-training large language models (LLMs). In practice, LLM RL is often off-policy because of training-inference mismatch and policy staleness, making trust-region control essential for stable optimization. Mainstream methods such as PPO and GRPO approximate this control with a ratio-clipping mechanism, but the importance ratio can be a poor proxy for distributional shift in long-tailed vocabularies. Recent work such as DPPO addresses this mismatch by replacing ratio-based clipping with a divergence-based mask, yielding a trust region defined by the sampled token's absolute probability shift. However, DPPO still relies on a hard mask: once a token crosses the trust-region boundary in a harmful direction, its gradient is discarded rather than corrected. To address this, we propose Divergence Regularized Policy Optimization (DRPO), which replaces the hard mask with a smooth advantage-weighted quadratic regularizer on policy shift. DRPO preserves the same trust-region geometry as DPPO while inducing bounded, continuous gradient weights that attenuate diverging updates and provide corrective signals beyond the boundary. Experiments across model scales, architectures, and precision settings show that DRPO improves the stability and efficiency of LLM RL training.

Jiarui Yao, Xiangxin Zhou, Penghui Qi +3

1Tencent Hunyuan · 2UIUC · 3NUS

RL for Language ModelsPolicy OptimizationTrust Region Policy OptimizationRL Post-TrainingOff-Policy RL
AskPDF
Jun 7, 2026

Structure-Conditioned Actor-Critic Branches for Quality-Diversity Reinforcement Learning

Quality-diversity reinforcement learning (QD-RL) aims to construct policy repertoires that contain both high-performing and behaviorally diverse policies. Existing QD-RL methods mainly diversify policy instances after rollout evaluation or use learned value information to improve policy quality and behavior targeting, while the learning branches that generate candidate policies remain less explored. This paper proposes SV-QD-RL, a structure-value coupled framework that represents each candidate as a structure-conditioned actor-critic branch. Each branch contains an actor, a structural mask, a branch-specific critic, a replay state, and evaluation attributes including behavior, return, sparsity, and value profile. The structural mask defines the actor subspace in which the branch learns, while the branch-specific critic and replay state shape its value-learning trajectory. A branch-aware QD archive then evaluates and retains branches according to behavioral quality, structural footprint, and value-profile information. Experiments on MuJoCo continuous-control tasks show that SV-QD-RL constructs policy repertoires with strong archive quality and behaviorally useful diversity. Ablation and diagnostic analyses further indicate that structural conditioning, critic differentiation, and memory-consistent refinement make complementary contributions to behavioral specialization. Schedule-aware repertoire evaluation shows that the learned archive provides selectable policy alternatives under changing behavior-level requirements. These results suggest that coupling actor structure with branch-specific value learning is an effective mechanism for generating diverse QD-RL policy repertoires.

Lianrong Zuo, Peilan Xu, Yong Liu +1

Guangdong Provincial Key Laboratory of Novel Security Intelligence Technologies, Institute of Cyberspace Security, School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen 518055, Guangdong, China

Reinforcement LearningPolicy OptimizationActor-Critic MethodsBehavioral Diversity in RL
AskPDF
Jun 6, 2026

The Easy, the Hard, and the Learnable: Confidence and Difficulty-Adaptive Policy Optimization for LLM Reasoning

RL with verifiable rewards can substantially improve LLM reasoning, yet standard GRPO-style training often treats easy, hard, and learnable questions alike through uniform sampling and weighting, leading to inefficient compute allocation. We study GRPO by tracking token log-probabilities, group-normalized advantages, and the induced token-level update weights. This reveals three recurring dynamics as training proceeds: (1) confidence inflation, (2) advantage contraction, and (3) hierarchical convergence. These findings suggest that the utility of each update depends strongly on both question difficulty and the model's current competence. Motivated by this, we propose Confidence and Difficulty-adaptive Policy Optimization (CoDaPO), which assigns each question a bounded value from rollout confidence and empirical difficulty. CoDaPO then uses this value to reweight policy updates and resample high-value learnable questions within mini-batches, thereby increasing discovery within the learnable band under a fixed compute budget. Across twelve benchmarks, CoDaPO consistently improves accuracy over existing RL methods. Our code is publicly available at https://github.com/tmlr-group/CoDaPO.

Zhanke Zhou, Xiangyu Lu, Chentao Cao +4

1TMLR Group, Department of Computer Science, Hong Kong Baptist University · 2Stanford University · 3Sydney AI Centre, The University of Sydney

RL for Language Model ReasoningPolicy OptimizationReinforcement Learning with Verifiable RewardsPolicy Gradient Methods
AskPDF
Jun 4, 2026

Retry Policy Gradients in Continuous Action Spaces

Retry-based objectives such as pass@K and max@K optimize the best return obtained from multiple sampled trajectories, and recent work has shown that they can promote exploration without explicit exploration bonuses. In discrete action spaces, ReMax was shown to do so by adapting to return uncertainty. In this work, we introduce pathwise derivative estimators for retry objectives and use them to extend ReMax to continuous action spaces. We study the resulting learning dynamics and show that, even with deterministic rewards, ReMax can encourage stochastic exploration by reshaping the policy-gradient landscape. In particular, it alters gradients both in direction, biasing updates toward higher policy entropy, and in magnitude, damping gradients and slowing convergence. We further show that Adam's adaptive normalization can mitigate this damping, depending on its numerical stabilization parameter. Empirically, we instantiate this objective as ReMax Actor-Critic (ReMAC), an off-policy actor--critic algorithm that optimizes the ReMax objective using a pathwise derivative estimator. Our experiments show that ReMAC can promote higher policy entropy without entropy regularization and achieves performance comparable to SAC.

Soichiro Nishimori, Paavo Parmas

The University of Tokyo, Japan

Policy OptimizationRL ExplorationPolicy Gradient MethodsActor-Critic MethodsPolicy Learning
AskPDF
Jun 3, 2026

Neetyabhas: A Framework for Uncertainty-Aware Public Policy Optimization in Rational Agent-Based Models

Purpose The WHO's COVID-19 non-pharmaceutical interventions (e.g., lockdowns, vaccinations) effectively curb transmission but impose heavy economic strains. Existing research often neglects individual behaviors and falsely assumes perfect infection tracking and flawless policy execution, failing to account for real-world uncertainties and errors. Methods We propose an integrative approach incorporating uncertainties in both epidemic measurement (infections/hospitalizations) and policy implementation. We built a simulation model of 1,000 individuals making real-time choices regarding mask-wearing, vaccination, and shopping. Concurrently, policymakers deploy interventions (lockdowns, mandates) based on health and economic observations. This framework is driven by hierarchical reinforcement learning agents, utilizing deep Q-networks alongside uncertainty-aware policy gradient variants (DDPG and TD3). Results The simulations effectively managed the epidemic's progression. Masking and vaccinations proved highly effective, significantly reducing both the outbreak's peak height and duration. By integrating individual behaviors, policy uncertainties, and multifaceted interventions, our dynamic control approach successfully mitigated the epidemic's impact. Conclusions Our model overcomes previous research limitations by embedding uncertainty and human behavior into public health policy frameworks. The simulation demonstrates that accounting for individual choices and imperfect data is crucial for designing effective interventions during complex pandemics, with masks and vaccines serving as pivotal tools.

Janani Venugopalan, Gaurav Deshkar, Rishabh Gaur +2

Engineering for Research (e4r), Thoughtworks Technologies, Binarius Building, Deepak Complex, Shastrinagar, Yerawada, Pune, Maharashtra, India - 411006

Policy OptimizationAgent-Based Modeling
AskPDF
Jun 2, 2026

Physics-Guided Policy Optimization with Self-Distillation

Self-distilled policy optimization (SDPO) has become a popular paradigm for LLM post-training, where a model learns from its own predictions conditioned on privileged information. SDPO, however, is sensitive to how much each update step should be trusted: corrections from a self-teacher can be highly informative on some batches and misleading on others, and applying them uniformly with a fixed step size can destabilize training. Drawing inspiration from viscous-fluid dynamics and formalizing the analogy at the SDE level, we propose Physics-Guided Policy Optimization (PGPO), which introduces an information-modulated step-size multiplier derived from a mutual-information estimate between the student's predictions and the feedback-conditioned teacher. We show that this modulation preserves the order-1 weak-approximation guarantees of vanilla SGD, and incurs negligible overhead per iteration. We evaluate PGPO on the Science-QA dataset, where it outperforms SDPO on 3 of the 4 domains with gains of up to +4.5 points, while remaining stable in a setting where SDPO collapses late in training.

Ke Wang, Yuning Wu, Haoran Liu +3

Amazon, Arlington, USA · Amazon, New York, USA · Amazon, San Jose, USA +1

On-Policy Self-DistillationPolicy OptimizationAdaptive Learning RateLLM Post-Training
AskPDF
Jun 1, 2026

From Global Policies to Local Strategies: Multi-Objective Optimization of Resource-Specific Handover Policies

Efficient resource allocation is a key challenge in business process management, with direct implications for cost, throughput time, and utilization. While recent Reinforcement Learning (RL) approaches have shown promise in deriving adaptive allocation policies, they typically neglect inter-resource collaboration patterns that can strongly influence real-world task handovers. Recognizing this, this paper introduces the first approach for multi-objective optimization of resource-level decision-making, enabling the recommendation of person-specific handover policies. To achieve this, our work combines an existing Multi-Agent System-based process simulator with a multi-objective evolutionary algorithm. The resulting approach produces Pareto-optimal, resource-specific policies that optimize the process across multiple objectives. Experimental results on synthetic and real-world datasets show that our approach reduces costs by an average of 37% and waiting time by 58%, consistently outperforming heuristic baselines and demonstrating the potential of leveraging collaboration-aware optimization to improve process performance.

Lukas Kirchdorfer, Artemis Doumeni, Han van der Aa +1

SAP Signavio, Walldorf, Germany · DTU Compute, Technical University of Denmark, Kongens Lyngby, Denmark · Faculty of Computer Science, University of Vienna, Vienna, Austria

Policy OptimizationMulti-Objective Optimization
AskPDF
May 30, 2026

Regularized Offline Policy Optimization with Posterior Hybrid Bayesian Belief

Offline reinforcement learning (RL) aims to optimize policies from pre-collected datasets. A bottleneck of this paradigm is managing epistemic uncertainty, which arises from limited data coverage (sample-level) and the ambiguity in identifying transition dynamics from finite data (model-level). To provide a unified quantification of these uncertainties, Bayesian RL has been proposed by treating the dynamics model as a random variable and maintaining a corresponding belief. Despite its theoretical appeal, policy optimization in Bayesian RL remains computationally challenging as it requires solving composite objectives with expectations. Prior methods either employ search-based techniques with poor computational scalability or impose restrictive posterior assumptions that sacrifice the adaptability of Bayesian RL. To address these limitations, we propose Posterior Hybrid Bayesian Belief (PhyB), which reformulates the expectation as a convex combination over a subset of dynamics models. Theoretical analysis demonstrates that the objective discrepancy induced by this approximation remains bounded. Based on PhyB, we develop an iterative regularized policy optimization algorithm that provides metric-agnostic guarantees for monotonic improvement until convergence. Empirical results demonstrate that PhyB achieves state-of-the-art performance on various benchmarks.

Hongqiang Lin, Pengfei Wang, Nenggan Zheng

Zhejiang University, Hangzhou, China.

Reinforcement LearningBayesian RLPolicy OptimizationOffline RLModel-Based RL
AskPDF
May 29, 2026

When are LLMs Sufficient Policy Optimizers for Sequential RL Tasks?

We study when large language models (LLMs) can serve as effective black-box policy optimizers for reinforcement learning (RL) tasks, i.e., when can we replace classical RL algorithms with an LLM? We explore this question by introducing Prompted Policy Optimization (PromptPO), an iterative method that prompts an LLM with Python descriptions of the state space, action space, and reward function, then has it generate and refine executable policies based on rollout feedback. Across hard exploration environments, Meta-World robotics tasks, and several real-world control problems, PromptPO often matches or exceeds the performance of standard RL baselines while using substantially fewer environment interactions. To maximize expected return, and without further explicit prompting, the policies PromptPO outputs range from tuned proportional controllers or rule-based plans to policies that run planning algorithms like value iteration. Our results demonstrate that LLM-based policy optimization is sufficient when the LLM can leverage prior knowledge about the environment or optimization strategy. PromptPO underperforms standard RL baselines in MuJoCo domains. This demonstrates possible limitations of LLM-based policy optimization to settings that requiring fine-grained continuous control.

Stephane Hatgis-Kessell, Emma Brunskill

Department of Computer Science, Stanford University

Language Model-Based ControlPolicy OptimizationLLM-Guided RLRobotic RLPolicy Learning
AskPDF
May 28, 2026

HPO: Hysteretic Policy Optimization for Stable and Efficient Training under Sparse-Reward Regime

We investigate a narrow but common failure mode of GRPO-style reinforcement learning in the context of sparse verifiable rewards: early updates contain more responses with negative advantages than those with positive advantages, while response-level length normalization ties the magnitude of the update to the length of the output. We propose Hysteretic Policy Optimization (HPO), a minimal modification of GRPO that reduces the weight of negative-advantage updates and replaces per-response length normalization with mean-length normalization. We further introduce Adaptive HPO (A-HPO), which sets the hysteretic weight based on batch-level advantage-sign statistics, thereby removing the need for tuning a fixed hysteretic weight. In our TeleLogs and Countdown experiments, A-HPO improves the reward per update compared to GRPO, with the largest gains in early sparse reward regimes. On TeleLogs, A-HPO achieves a final reward of 0.84, outperforming SAPO by 5%, GSPO by 11%, and GRPO by 15%, while maintaining a comparable response-length. On Countdown, A-HPO achieves the largest gains in initial and most difficult configurations across 1.5B-7B models. Ablation studies on the hysteretic weight show that the gains of A-HPO come from better balancing the contributions of positive and negative advantages compared to positive-only or fully symmetric updates.

Mohamed Sana, Nicola Piovesan, Antonio De Domenico +2

Paris Research Center, Huawei Technologies, Boulogne-Billancourt, France · Huawei Technologies, China

Group Relative Policy OptimizationPolicy OptimizationReinforcement Learning with Verifiable RewardsRL Post-TrainingSparse-Reward RLPolicy Learning
AskPDF
May 28, 2026

Guidance Contrastive Token Credit Assignment for Discrete Policy Optimization

Group-advantage-based reinforcement learning methods, such as GRPO and DAPO, have demonstrated strong performance across diverse domains, including mathematical reasoning and text-to-image generation. However, their reliance on sample-level rewards introduces a key limitation as uniform credit assignment across all tokens fails to capture fine-grained, token-level contributions. To address this issue, we propose Guidance Contrastive Policy Optimization (GCPO), a novel algorithm that enables per-token credit assignment by contrasting model predictions under positive and negative prompts. Rather than uniformly broadcasting sample-level advantages, GCPO assigns token-level advantages proportional to the difference between these contrastive predictions, allowing more precise and informative learning signals. Empirically, we find that GCPO emphasizes semantically relevant regions such as visual areas aligned with textual prompts in text-to-image generation, and critical keywords within reasoning traces for chain-of-thought tasks. Through extensive experiments, GCPO consistently outperforms GRPO and DAPO baselines on both text-to-image generation and chain-of-thought reasoning benchmarks, demonstrating its effectiveness as a general and scalable optimization strategy for discrete policy learning.

Shufan Li, Konstantinos Kallidromitis, Akash Gokul +2

UCLA · Panasonic AI Research · NVIDIA

Token-Level Credit AssignmentRL for Language Model ReasoningCredit Assignment in RLPolicy OptimizationRL for Image GenerationPolicy Gradient Methods
AskPDF
May 27, 2026

Insurance Pricing Optimization via Off-Policy Evaluation

Traditional insurance pricing relies on risk-based principles that ensure actuarial fairness and solvency but do not explicitly account for policyholders' price sensitivity. We formulate insurance pricing as a decision-making problem and study it using tools from off-policy evaluation and stochastic control. We propose a kernelized inverse propensity score estimator that exploits local structure in the action space and yields variance reduction compared to the classical inverse propensity score estimator. Building on these value estimates, we investigate policy optimization and present two practical approaches for computing optimal pricing rules: an interpretable data-shared Lasso formulation and a flexible policy parameterization based on neural networks. Using a controlled synthetic travel insurance environment, we empirically confirm the theoretical results and show that neural networks outperform existing techniques for policy optimization.

Sascha Günther, Dimitri Semenovich, Mario V. Wüthrich

Department of Mathematics, ETH Zurich

Off-Policy EvaluationPolicy Optimization
AskPDF
May 27, 2026

Long Live The Balance: Information Bottleneck Driven Tree-based Policy Optimization

Recent advances in online reinforcement learning (RL) for large language models (LLMs) have demonstrated promising performance in complex reasoning tasks. However, they often exhibit an imbalanced exploration-exploitation trade-off, resulting in unstable optimization and sub-optimal performance. We introduce IB-Score, a novel metric grounded in Information Bottleneck theory that evaluates policy's exploration-exploitation balance by quantifying the trade-off between step-level reasoning diversity and mutual information shared with the correct answer. Analysis based on IB-Score shows that popular online RL approaches (e.g., GRPO) with common regularizers fail to consistently maintain balance during training with suboptimal results. To address this, we propose Information Bottleneck-driven Tree-based Policy Optimization (IB-TPO), a principled framework that formulates IB-Score as a fine-grained optimization objective and utilizes a novel IB-guided tree sampling strategy that not only improves the efficiency of online sampling with 50% more trajectories under the same token budget, but also reuses the tree structure for effective IB-Score Monte Carlo estimation. Extensive experiments across standard benchmarks show that our method significantly outperforms GRPO baseline by 2.9% to 3.6% and also outperforms other state-of-the-art online RL approaches. Our code is available at https://github.com/alibaba/EfficientRL.

Hao Jiang, Shurui Li, Tianpeng Bu +7

Alibaba Cloud Computing, Alibaba Group.

RL for Language ModelsExploration-Exploitation TradeoffRL for Language Model ReasoningInformation BottleneckPolicy Optimization
AskPDF
May 26, 2026

Ratio-Variance Regularized Policy Optimization

Standard on-policy reinforcement learning relies on heuristic clipping to enforce trust regions, but this mechanism imposes a severe cost by indiscriminately truncating high-return yet high-divergence updates. We demonstrate that explicitly constraining the policy ratio variance provides a principled local approximation to trust-region constraints, eliminating the need for binary hard clipping. By acting as a distributional ``soft brake'', this approach preserves critical gradient signals from novel discoveries while naturally down-weighting and enabling the reuse of stale, off-policy data. We introduce R2VPO{\bf R}^2{\bf VPO} (Ratio-Variance Regularized Policy Optimization), which implements this constraint via a primal-dual optimization framework. Extensive evaluations across 77 LLM scales, spanning both fast and slow reasoning paradigms, and 1010 robotic control tasks demonstrate the generality of the proposed approach. R2^2VPO achieves substantial performance gains on mathematical reasoning benchmarks, with particularly pronounced improvements on smaller models, while significantly improving sample efficiency. Furthermore, it consistently outperforms PPO baselines in continuous control domains, particularly in sparse-reward and dynamic environments. Together, these findings establish ratio-variance regularization as a principled foundation for stable and data-efficient policy optimization.

Yu Luo, Shuo Han, Yihan Hu +5

Department of Foundation Model, 2012 Labs, Huawei Shanghai · Shanghai Research Institute for Intelligent Autonomous Systems, Tongji University · Department of Computer Science and Technology, Tsinghua University +1

RL for Language Model ReasoningPolicy OptimizationRobotic RLProximal Policy OptimizationSample-Efficient RLRegularized RL
AskPDF
May 26, 2026

AgensFlow: A Coordination-Policy Substrate for Multi-Agent Systems

Multi-agent systems built on large language models (LLMs) require many coordination choices that are difficult to fix a priori: which skill protocol to invoke, which agent role should perform a subtask, which model to bind to each role, how roles should interact, when to use retrieval or verification, and when to omit a step entirely. These choices interact with task regime and operational constraints, so static pipelines and one-off model comparisons provide only a limited view of the design space. This paper introduces AgensFlow, an open-source framework that treats multi-agent coordination as an online policy-learning problem under partial observability. The framework makes coordination decisions observable and learnable from repeated trajectories, rather than treating skill, role, model, topology, and evaluation choices as fixed pipeline design. AgensFlow is evaluated on two corpora: distributed-systems incident tasks and security-advisory tasks. The evaluation shows three main results: learned routing reaches a higher-quality operating point than a fixed pipeline baseline on coordination-heavy classes; skip:X isolates topology compression as a meaningful part of the substrate; and warm-started policy graphs can reduce exploration cost while preserving plateau quality. Overall, the results support that learned, auditable routing can improve coordination-heavy multi-agent workflows over static wiring.

Nicole Koenigstein

Independent researcher.

Multi-Agent LLM SystemsMulti-Agent System OptimizationMulti-Agent CoordinationLLM Agent OrchestrationPolicy Optimization
AskPDF