Policy Learning

Follow this topic Ask about this topic

Policy learning studies how to choose actions or interventions over time to improve rewards, welfare, or other goals, often when outcomes are uncertain or feedback is limited. Common approaches include policy-gradient and actor–critic reinforcement learning, value-based and offline learning, imitation from demonstrations, and direct learning of treatment or resource-allocation rules. Current work also tackles long-horizon language-model agents and generative robot policies, using methods such as finer-grained credit assignment, exploration, memory, and policy reuse. A central tension is achieving reliable, safe learning with theoretical guarantees while using expressive policies and imperfect data or rewards efficiently.

Latest papers

488 papers74 papers in the last four weeks, up 48% on the four weeks before. 0.4% of all new papers.

Apr 16, 2026

Mean Flow Policy Optimization

Diffusion models have recently emerged as expressive policy representations for online reinforcement learning (RL). However, their iterative generative processes introduce substantial training and inference overhead. To overcome this limitation, we propose to represent policies using MeanFlow models, a class of few-step flow-based generative models, to improve training and inference efficiency over diffusion-based RL approaches. To promote exploration, we optimize MeanFlow policies under the maximum entropy RL framework via soft policy iteration, and address two key challenges specific to MeanFlow policies: action likelihood evaluation and soft policy improvement. Experiments on MuJoCo, DeepMind Control Suite and HumanoidBench benchmarks demonstrate that our method, Mean Flow Policy Optimization (MFPO), achieves performance comparable to or exceeding current diffusion-based baselines while considerably reducing training and inference time. Our code is available at https://github.com/dongxiaoyi-xyz/MFPO.

Xiaoyi Dong, Xi Sheryl Zhang, Jian Cheng

C2DL, Institute of Automation, Chinese Academy of Sciences · School of Artificial Intelligence, University of Chinese Academy of Sciences · AiRiA +1

MeanFlow Mean Flow Continuous Control Policy Learning Flow-Based Policy Learning

Apr 15, 2026

When Missing Becomes Structure: Intent-Preserving Policy Completion from Financial KOL Discourse

Key Opinion Leader (KOL) discourse on social media is widely consumed as investment guidance, yet turning it into executable trading strategies without injecting assumptions about unspecified execution decisions remains an open problem. We observe that the gaps in KOL statements are not random deficiencies but a structured separation: KOLs express directional intent (what to buy or sell and why) while leaving execution decisions (when, how much, how long) systematically unspecified. Building on this observation, we propose an intent-preserving policy completion framework that treats KOL discourse as a partial trading policy and uses offline reinforcement learning to complete the missing execution decisions around the KOL-expressed intent. Experiments on multimodal KOL discourse from YouTube and X (2022-2025) show that KICL achieves the best return and Sharpe ratio on both platforms while maintaining zero unsupported entries and zero directional reversals, and ablations confirm that the full framework yields an 18.9% return improvement over the KOL-aligned baseline.

Yuncong Liu, Yuan Wan, Zhou Jiang +1

Asian Institute of Digital Finance, National University of Singapore · School of Computing, National University of Singapore

Algorithmic Trading Policy Learning

Apr 7, 2026

HiPolicy: Hierarchical Multi-Frequency Action Chunking for Policy Learning

Robotic imitation learning faces a fundamental trade-off between modeling long-horizon dependencies and enabling fine-grained closed-loop control. Existing fixed-frequency action chunking approaches struggle to achieve both. Building on this insight, we propose HiPolicy, a hierarchical multi-frequency action chunking framework that jointly predicts action sequences at different frequencies to capture both coarse high-level plans and precise reactive motions. We extract and fuse hierarchical features from history observations aligned to each frequency for multi-frequency chunk generation, and introduce an entropy-guided execution mechanism that adaptively balances long-horizon planning with fine-grained control based on action uncertainty. Experiments on diverse simulated benchmarks and real-world manipulation tasks show that HiPolicy can be seamlessly integrated into existing 2D and 3D generative policies, delivering consistent improvements in performance while significantly enhancing execution efficiency.

Jiyao Zhang, Zimu Han, Junhan Wang +7

CFCS, School of CS, PKU · National Key Laboratory for Multimedia Information Processing, School of CS, PKU · XJTU +3

Policy Learning Action Chunking Robot Imitation Learning Long-Horizon Robotic Manipulation Hierarchical RL

Apr 5, 2026

MC-CPO: Mastery-Conditioned Constrained Policy Optimization for Pedagogically Safe Intelligent Tutoring Systems

Intelligent tutoring systems increasingly rely on reinforcement learning to personalise instruction, yet optimising for observable engagement signals can systematically decouple learner activity from genuine knowledge acquisition. Analysing over 21 million student interactions across two deployed platforms, we find engagement events without corresponding mastery gains occur in 26.5% of interactions on Junyi Academy (72,758 students) and 3.1% on XES3G5M (14,453 students, NeurIPS 2023), confirming this pattern is directly observable in deployed educational technology at scale. We introduce Mastery-Conditioned Constrained Policy Optimisation (MC-CPO), a reinforcement learning framework that addresses this problem structurally. MC-CPO conditions the admissible instructional action space on learner mastery state: a concept becomes available only when prerequisite knowledge meets a mastery threshold, yielding an action space that expands naturally as learners acquire knowledge. Pedagogical safety constraints are enforced by construction, with formal guarantees of structural prerequisite safety, primal-dual convergence, and strict dominance over post-hoc filtering. MC-CPO is the only method to reduce reward hacking severity across all conditions. Mean per-episode mastery gain increases by 18.3% on Junyi Academy and 54.0% on XES3G5M relative to all baselines, while competitive engagement performance is maintained. These results support structural constraint modelling as a principled foundation for safer adaptive instructional policies in deployed tutoring systems.

Oluseyi Olukola, Nick Rahimi

School of Computing Sciences and Computer Engineering University of Southern Mississippi Hattiesburg MS 39406 USA

Intelligent Tutoring Systems Policy Learning Safe RL Constrained RL Reward Hacking

Apr 1, 2026

Model-Based Learning of Near-Optimal Finite-Window Policies in POMDPs

We study model-based learning of finite-window policies in tabular partially observable Markov decision processes (POMDPs). A common approach to learning under partial observability is to approximate unbounded history dependencies using finite action-observation windows. This induces a finite-state Markov decision process (MDP) over histories, referred to as the superstate MDP. Once a model of this superstate MDP is available, standard MDP algorithms can be used to compute optimal policies, motivating the need for sample-efficient model estimation. Estimating the superstate MDP model is challenging because trajectories are generated by interaction with the original POMDP, creating a mismatch between the sampling process and target model. We propose a model estimation procedure for tabular POMDPs and analyze its sample complexity. Our analysis exploits a connection between filter stability and concentration inequalities for weakly dependent random variables. As a result, we obtain tight sample complexity guarantees for estimating the superstate MDP model from a single trajectory. Combined with value iteration, this yields approximately optimal finite-window policies for the POMDP.

Philip Jordan, Maryam Kamgarpour

SYCAMORE group, Institute of Mechanical Engineering, EPFL Switzerland.

Partially Observable Markov Decision Processes Policy Learning Model-Based RL Sample-Efficient RL

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 30, 2026

Multi-Agent LLMs for Adaptive Acquisition in Bayesian Optimization

The exploration-exploitation trade-off is central to sequential decision-making and black-box optimization, yet how Large Language Models (LLMs) reason about and manage this trade-off remains poorly understood. Unlike Bayesian Optimization, where exploration and exploitation are explicitly encoded through acquisition functions, LLM-based optimization relies on implicit, prompt-based reasoning over historical evaluations, making search behavior difficult to analyze or control. In this work, we present a metric-level study of LLM-mediated search policy learning, studying how LLMs construct and adapt exploration-exploitation strategies under multiple operational definitions of exploration, including informativeness, diversity, and representativeness. We show that single-agent LLM approaches, which jointly perform strategy selection and candidate generation within a single prompt, suffer from cognitive overload, leading to unstable search dynamics and premature convergence. To address this limitation, we propose a multi-agent framework that decomposes exploration-exploitation control into strategic policy mediation and tactical candidate generation. A strategy agent assigns interpretable weights to multiple search criteria, while a generation agent produces candidates conditioned on the resulting search policy defined as weights. This decomposition renders exploration-exploitation decisions explicit, observable, and adjustable. Empirical results across various continuous optimization benchmarks indicate that separating strategic control from candidate generation substantially improves the effectiveness of LLM-mediated search.

Mohammadsina Almasi, Andrea Carbonati, Hadis Anahideh

University of Illinois Chicago, IL, USA

Large Language Model-Guided Optimization Exploration-Exploitation Tradeoff Multi-Agent LLMs Policy Learning Bayesian Optimization

Mar 30, 2026

Functional Natural Policy Gradients

We propose a cross-fitted debiasing device for policy learning from offline data. A key consequence of the resulting learning principle is

\sqrt N

regret even for policy classes with complexity greater than Donsker, provided a product-of-errors nuisance remainder is

O(N^{-1/2})

. The regret bound factors into a plug-in policy error factor governed by policy-class complexity and an environment nuisance factor governed by the complexity of the environment dynamics, making explicit how one may be traded against the other.

Aurelien Bibaut, Houssam Zenati, Thibaud Rahier +1

Cross-Fitting Natural Policy Gradient Policy Learning Offline RL Regret Minimization in RL Policy Gradient

Mar 30, 2026

Optimistic Actor-Critic with Parametric Policies for Linear Markov Decision Processes

Although actor-critic methods have been successful in practice, their theoretical analyses have several limitations. Specifically, existing theoretical work either sidesteps the exploration problem by making strong assumptions or analyzes impractical methods with complicated algorithmic modifications. Moreover, the actor-critic methods analyzed for linear MDPs often employ natural policy gradient and construct "implicit" policies without explicit parameterization. Such policies are computationally expensive to sample from, making the environment interactions inefficient. To that end, we focus on the finite-horizon linear MDPs and propose an optimistic actor-critic framework that uses parametric log-linear policies. In particular, we introduce a tractable

\textit{logit-matching}

regression objective for the actor. For the critic, we use approximate Thompson sampling via Langevin Monte Carlo to obtain optimistic value estimates. We prove that the resulting algorithm achieves

\widetilde{\mathcal{O}}(ε^{-4})

and

\widetilde{\mathcal{O}}(ε^{-2})

sample complexity in the on-policy and off-policy setting, respectively. Our results match prior theoretical work in achieving the state-of-the-art sample complexity, while our algorithm is more aligned with practice.

Max Qiushi Lin, Reza Asad, Kevin Tan +3

Simon Fraser University · University of Pennsylvania · Mila, McGill University +1

Actor-Critic Methods Policy Learning Linear Markov Decision Processes Sample-Efficient RL

Mar 25, 2026

Towards Safe Learning-Based Non-Linear Model Predictive Control through Recurrent Neural Network Modeling

The practical deployment of nonlinear model predictive control (NMPC) is often limited by online computation: solving a nonlinear program at high control rates can be expensive on embedded hardware, especially when models are complex or horizons are long. Learning-based NMPC approximations shift this computation offline but typically demand large expert datasets and costly training. We propose Sequential-AMPC, a sequential neural policy that generates MPC candidate control sequences by sharing parameters across the prediction horizon. For deployment, we wrap the policy in a safety-augmented online evaluation and fallback mechanism, yielding Safe Sequential-AMPC. Compared to a naive feedforward policy baseline across several benchmarks, Sequential-AMPC requires substantially fewer expert MPC rollouts and yields candidate sequences with higher feasibility rates and improved closed-loop safety. On high-dimensional systems, it also exhibits better learning dynamics and performance in fewer epochs while maintaining stable validation improvement where the feedforward baseline can stagnate.

Mihaela-Larisa Clement, Mónika Farsang, Agnes Poks +4

Institute of Computer Engineering, TU Wien, Vienna, Austria · AIT Austrian Institute of Technology, Vienna, Austria · Institute of Mechanics and Mechatronics, TU Wien, Vienna, Austria

Nonlinear MPC Model Predictive Control Policy Learning Safety-Critical Control Recurrent Neural Networks

Mar 24, 2026

End-to-End Efficient RL for Linear Bellman Complete MDPs with Deterministic Transitions

We study reinforcement learning (RL) with linear function approximation in Markov Decision Processes (MDPs) satisfying linear Bellman completeness -- a fundamental setting where the Bellman backup of any linear value function remains linear. While statistically tractable, prior computationally efficient algorithms are either limited to small action spaces or require strong oracle assumptions over the feature space. We provide a computationally efficient algorithm for linear Bellman complete MDPs with deterministic transitions, stochastic initial states, and stochastic rewards. For finite action spaces, our algorithm is end-to-end efficient; for large or infinite action spaces, we require only a standard argmax oracle over actions. Our algorithm learns an

\varepsilon

-optimal policy with sample and computational complexity polynomial in the horizon, feature dimension, and

1/\varepsilon

.

Zakaria Mhammedi, Alexander Rakhlin, Nneka Okolo

Google Research, NYC · MIT

Reinforcement Learning Linear Markov Decision Processes Policy Learning Sample-Efficient RL Function Approximation in RL

Mar 20, 2026

Delightful Distributed Policy Gradient

Distributed reinforcement learning trains on data from stale, buggy, or mismatched actors, producing actions with high surprisal (negative log-probability) under the learner's policy. The core difficulty is not surprising data per se, but negative learning from surprising data. High-surprisal failures can dominate finite-batch updates through large perpendicular components, while high-surprisal successes reveal opportunities the current policy would otherwise miss. The Delightful Policy Gradient (DG) separates these cases by gating each update with delight, the product of advantage and surprisal, suppressing rare failures and preserving rare successes without behavior probabilities. In a tabular analysis, DG suppresses the perpendicular second moment of high-surprisal failures by a policy-overlap factor that vanishes as the learner improves. The advantage sign is essential for surprisal-based filtering: any learner-probability-only gate that suppresses rare failures also suppresses rare successes. On MNIST with simulated staleness, DG without off-policy correction outperforms importance-weighted PG with exact behavior probabilities. On a transformer sequence task with staleness, actor bugs, reward corruption, and rare discovery, DG often achieves nearly order-of-magnitude lower error. When all four frictions act simultaneously, its sample-efficiency advantage is order-of-magnitude and grows with task complexity.

Ian Osband

Google DeepMind

Policy Gradient Methods Policy Learning Distributed RL Sample-Efficient RL Off-Policy RL

Mar 20, 2026

Investigating a Policy-Based Formulation for Endoscopic Camera Pose Recovery

In endoscopic surgery, surgeons continuously locate the endoscopic view relative to the anatomy by interpreting the evolving visual appearance of the intraoperative scene in the context of their prior knowledge. Vision-based navigation systems seek to replicate this capability by recovering camera pose directly from endoscopic video, but most approaches do not embody the same principles of reasoning about new frames that makes surgeons successful. Instead, they remain grounded in feature matching and geometric optimization over keyframes, an approach that has been shown to degrade under the challenging conditions of endoscopic imaging like low texture and rapid illumination changes. Here, we pursue an alternative approach and investigate a policy-based formulation of endoscopic camera pose recovery that seeks to imitate experts in estimating trajectories conditioned on the previous camera state. Our approach directly predicts short-horizon relative motions without maintaining an explicit geometric representation at inference time. It thus addresses, by design, some of the notorious challenges of geometry-based approaches, such as brittle correspondence matching, instability in texture-sparse regions, and limited pose coverage due to reconstruction failure. We evaluate the proposed formulation on cadaveric sinus endoscopy. Under oracle state conditioning, we compare short-horizon motion prediction quality to geometric baselines achieving lowest mean translation error and competitive rotational accuracy. We analyze robustness by grouping prediction windows according to texture richness and illumination change indicating reduced sensitivity to low-texture conditions. These findings suggest that a learned motion policy offers a viable alternative formulation for endoscopic camera pose recovery.

Jan Emily Mangulabnan, Akshat Chauhan, Laura Fleig +7

Johns Hopkins University, Baltimore MD 21218, USA · Johns Hopkins Medical Institutions, Baltimore MD 21218, USA

Camera Pose Estimation Visual Odometry Policy Learning Surgical Robotics

Mar 20, 2026

DeepStock: Reinforcement Learning with Policy Regularizations for Inventory Management

Deep Reinforcement Learning (DRL) provides a general-purpose methodology for training inventory policies that can leverage big data and compute. However, off-the-shelf implementations of DRL have seen mixed success, often plagued by high sensitivity to the hyperparameters used during training. In this paper, we show that by imposing policy regularizations, grounded in classical inventory concepts such as "Base Stock", we can significantly accelerate hyperparameter tuning and improve the final performance of several DRL methods. We report details from a 100% deployment of DRL with policy regularizations on Alibaba's e-commerce platform, Tmall. We also include extensive synthetic experiments, which show that policy regularizations reshape the narrative on what is the best DRL method for inventory management.

Yaqi Xie, Xinru Hao, Jiaxi Liu +4

Booth School of Business, University of Chicago, Chicago, USA. · Taobao & Tmall Group, Hangzhou, China. · School of Economics, Sichuan University, Chengdu, China. +2

Inventory Control Reinforcement Learning Policy Learning Regularized RL

Mar 19, 2026

Near-Equivalent Q-learning Policies for Dynamic Treatment Regimes

Precision medicine aims to tailor therapeutic decisions to individual patient characteristics. This objective is commonly formalized through dynamic treatment regimes, which use statistical and machine learning methods to derive sequential decision rules adapted to evolving clinical information. In most existing formulations, these approaches produce a single optimal treatment at each stage, leading to a unique decision sequence. However, in many clinical settings, several treatment options may yield similar expected outcomes, and focusing on a single optimal policy may conceal meaningful alternatives. We extend the Q-learning framework for retrospective data by introducing a worst-value tolerance criterion controlled by a hyperparameter

\varepsilon

, which specifies the maximum acceptable deviation from the optimal expected value. Rather than identifying a single optimal policy, the proposed approach constructs sets of

\varepsilon

-optimal policies whose performance remains within a controlled neighborhood of the optimum. This formulation shifts Q-learning from a vector-valued representation to a matrix-valued one, allowing multiple admissible value functions to coexist during backward recursion. The approach yields families of near-equivalent treatment strategies and explicitly identifies regions of treatment indifference where several decisions achieve comparable outcomes. We illustrate the framework in two settings: a single-stage problem highlighting indifference regions around the decision boundary, and a multi-stage decision process based on a simulated oncology model describing tumor size and treatment toxicity dynamics.

Sophia Yazzourh, Erica E. M. Moodie

Department of Epidemiology and Biostatistics McGill University Montréal, Québec, Canada

Q-Learning Policy Learning Sequential Decision Making

Mar 16, 2026

Generalizable Optimal Control with Transformers: One Policy Across Diverse Systems

Classical optimal control designs a separate controller for each plant. Even for the Linear Quadratic Regulator (LQR), every new model must be identified and its Riccati equation re-solved. We ask whether a single learned policy can instead serve an entire family of systems, and we show that one transformer can. We train the policy to imitate optimal LQR state feedback across a collection of heterogeneous Multiple-Input, Multiple-Output (MIMO) Linear Time-Invariant (LTI) systems that differ in their state and input dimensions and in their cost objectives. A shared representation lets the same parameters control every member of the family. It combines system-wise standardization, zero-padding and masking across dimensions, and an explicit encoding of the cost matrices. At run time, the policy maps a short window of recent states and the specified cost to a control action. It uses no plant matrices and identifies the dynamics implicitly from the state history. We evaluate on

28

simulated systems over

9{,}675

closed-loop rollouts, and no unstable rollout was observed in any of them. On the systems seen during training, it attains a median relative sub-optimality of

0.022\%

, even under parameter perturbations of up to

\pm10\%

. It transfers to unseen systems with lightweight fine-tuning, reaching a median sub-optimality of

0.19\%

. These results support transformers as generalizable near-optimal controllers for structured families of linear systems.

Turki Bin Mohaya, Maitham F. AL-Sunni, John M. Dolan +1

Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, USA

Optimal Control Neural Network Control Policy Learning

Mar 10, 2026

From Prior to Pro: Efficient Skill Mastery via Distribution Contractive RL Finetuning

We introduce Distribution Contractive Reinforcement Learning (DICE-RL), a framework that uses reinforcement learning (RL) as a "distribution contraction" operator to refine pretrained generative robot policies. DICE-RL turns a pretrained behavior prior into a high-performing "pro" policy by amplifying high-success behaviors from online feedback. We pretrain a diffusion- or flow-based policy for broad behavioral coverage, then finetune it with a stable, sample-efficient residual off-policy RL framework that combines selective behavior regularization with value-guided action selection. Extensive experiments and analyses show that DICE-RL reliably improves performance with strong stability and sample efficiency. It enables mastery of complex long-horizon manipulation skills directly from high-dimensional pixel inputs, both in simulation and on a real robot. Project website: https://zhanyisun.github.io/dice.rl.2026/.

Zhanyi Sun, Shuran Song

1Stanford University.

RL for Robotics Robot Policy Learning Robot Skill Learning Data-Efficient Robot Learning Residual Policy Learning Policy Learning+2

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

Vector Field Augmented Differentiable Policy Learning for Vision-Based Drone Racing

Autonomous drone racing in complex environments requires agile, high-speed flight while maintaining reliable obstacle avoidance. Differentiable-physics-based policy learning has recently demonstrated high sample efficiency and remarkable performance across various tasks, including agile drone flight and quadruped locomotion. However, applying such methods to drone racing remains difficult, as key objective like gate traversal are inherently hard to express as smooth, differentiable losses. To address these challenges, we propose DiffRacing, a novel vector field-augmented differentiable policy learning framework. DiffRacing integrates differentiable losses and vector fields into the training process to provide continuous and stable gradient signals, balancing obstacle avoidance and high-speed gate traversal. In addition, a differentiable Delta Action Model compensates for dynamics mismatch, enabling efficient sim-to-real transfer without explicit system identification. Extensive simulation and real-world experiments demonstrate that DiffRacing achieves superior sample efficiency, faster convergence, and robust flight performance, thereby demonstrating that vector fields can augment traditional gradient-based policy learning with a task-specific geometric prior.

Yang Su, Feng Yu, Yu Hu +4

The authors are with the Shanghai Jiao Tong University, Shanghai 200240, China

Autonomous Drone Racing Policy Learning Visuomotor Policy Learning

Mar 9, 2026

EnergyAction: Unimanual to Bimanual Composition with Energy-Based Models

Recent advances in unimanual manipulation policies have achieved remarkable success across diverse robotic tasks through abundant training data and well-established model architectures. However, extending these capabilities to bimanual manipulation remains challenging due to the lack of bimanual demonstration data and the complexity of coordinating dual-arm actions. Existing approaches either rely on extensive bimanual datasets or fail to effectively leverage pre-trained unimanual policies. To address this limitation, we propose EnergyAction, a novel framework that compositionally transfers unimanual manipulation policies to bimanual tasks through the Energy-Based Models (EBMs). Specifically, our method incorporates three key innovations. First, we model individual unimanual policies as EBMs and leverage their compositional properties to compose left and right arm actions, enabling the fusion of unimanual policies into a bimanual policy. Second, we introduce an energy-based temporal-spatial coordination mechanism through energy constraints, ensuring the generated bimanual actions are both temporal coherence and spatial feasibility. Third, we propose two different energy-aware denoising strategies that dynamically adapt denoising steps based on action quality assessment. These strategies ensure the generation of high-quality actions while maintaining superior computational efficiency compared to fixed-step denoising approaches. Experimental results demonstrate that EnergyAction effectively transfers unimanual knowledge to bimanual tasks, achieving superior performance on both simulated and real-world tasks with minimal bimanual data.

Mingchen Song, Xiang Deng, Jie Wei +3

Harbin Institute of Technology (Shenzhen) · PengCheng Laboratory · Shenzhen Loop Area Institute

Bimanual Robotic Manipulation Policy Learning Transfer Learning Energy-Based Models

Mar 6, 2026

Reinforcing the World's Edge: A Continual Learning Problem in the Multi-Agent-World Boundary

Ordinary decentralized multi-agent reinforcement learning presents each focal agent with a continual learning problem: peer updates change its induced rewards and dynamics even when the joint Markov game is stationary. We connect the lifetime of success-conditioned reusable structure to peer learning and policy reuse. An invariant core represents maximal abstract patterns shared by a high fraction of successful trajectories; survival refers to a fixed pattern's coverage, not an unchanged maximal frontier. An established sharp conditioning bound limits coverage loss to

\frac{\varepsilon}{p_0}

, where

\varepsilon

is trajectory-law drift,

p_0

is reference success mass, and drift is smaller than the initial compatible-success mass. Combining this bound with peer-policy movement certifies

Ω(\frac{1}η)

survival under bounded peer updates of size

η

and positive coverage margin. An effective-conflict condition yields a matching

Θ(\frac{1}η)

first-exit law and holds in an analytic exact-policy-gradient class. With success-mass and performance calibration, structural survival also yields policy-value and finite-library transfer guarantees. An exactly solvable corridor tests the structural predictions. Two registered 64-stream studies in continual control and cue-MNIST show that coverage erosion predicts impending failure and enables near-oracle intervention. An exploratory reanalysis of eight learned-partner Level-Based Foraging pairings suggests the same erosion--failure link under peer learning.

Dane Malenfant

School of Computer Science McGill University Mila - The Québec AI Institute

Non-Stationary RL Decentralized MARL Policy Learning Multi-Agent Reinforcement Learning Continual RL

Mar 5, 2026

Learning Optimal Individualized Decision Rules with Conditional Demographic Parity

Individualized decision rules (IDRs) have become increasingly prevalent in societal applications such as personalized marketing, healthcare, and public policy design. However, a critical ethical concern arises from the potential discriminatory effects of IDRs trained on biased data. These algorithms may disproportionately harm individuals from minority subgroups defined by sensitive attributes like gender, race, or language. To address this issue, we propose a novel framework that incorporates demographic parity (DP) and conditional demographic parity (CDP) constraints into the estimation of optimal IDRs. We show that the theoretically optimal IDRs under DP and CDP constraints can be obtained by applying perturbations to the unconstrained optimal IDRs, enabling a computationally efficient solution. Theoretically, we derive convergence rates for both policy value and the fairness constraint term. The effectiveness of our methods is illustrated through comprehensive simulation studies and an empirical application to the Oregon Health Insurance Experiment.

Wenhai Cui, Wen Su, Donglin Zeng +1

Department of Applied Mathematics, The Hong Kong Polytechnic University · Department of Biostatistics, City University of Hong Kong · Department of Biostatistics, University of Michigan

Algorithmic Fairness Policy Learning Group Fairness

Mar 3, 2026

RAPO: Expanding Exploration for LLM Agents via Retrieval-Augmented Policy Optimization

Agentic Reinforcement Learning (Agentic RL) has shown remarkable potential in large language model-based (LLM) agents. These works can empower LLM agents to tackle complex tasks via multi-step, tool-integrated reasoning. However, an inherent limitation of existing Agentic RL methods is their reliance on a pure on-policy paradigm for exploration, restricting exploration to the agent's self-generated outputs and preventing the discovery of new reasoning perspectives for further improvement. While recent efforts incorporate auxiliary off-policy signals to enhance exploration, they typically utilize full off-policy trajectories for trajectory-level policy estimation, overlooking the necessity for the fine-grained, step-level exploratory dynamics within agentic rollout. In this paper, we revisit exploration in Agentic RL and propose Retrieval-Augmented Policy Optimization (RAPO), a novel RL framework that introduces retrieval to explicitly expand exploration during training. To achieve this, we decompose the Agentic RL training process into two phases: (i) Hybrid-policy Agentic Rollout, and (ii) Retrieval-aware Policy Optimization. Specifically, we propose a Hybrid-policy Agentic Rollout strategy, which allows the agents to continuously reason over the retrieved off-policy step-level traces. It dynamically extends the reasoning receptive field of agents, enabling broader exploration conditioned on external behaviors. Subsequently, we introduce the Retrieval-aware Policy Optimization mechanism, which calibrates the policy gradient estimation with retrieval reward and importance shaping, stabilizing training and prioritizing retrieval-illuminating exploration. Extensive experiments show that RAPO achieves an +5.0% average gain on fourteen datasets across three agentic reasoning tasks, while delivering 1.2x faster training efficiency.

Siwei Zhang, Yun Xiong, Xi Chen +4

Fudan University · Zhejiang University · UC Davis

Agentic RL RL for Language Model Reasoning RL Exploration Policy Learning Off-Policy RL

Mar 3, 2026

Temporal Consistency Improves Generalization in Contextual Offline Meta Reinforcement Learning

Offline meta-reinforcement learning seeks to learn a policy that generalizes to new related tasks online. Context-based methods infer a task representation from transition histories, yet learning an effective task representation without supervision remains challenging. Existing methods relying on contrastive learning learn discriminative task representations, but fail to identify task-specific dynamics, while relying on reconstruction can be insufficient to model long-horizon dependencies, limiting generalization to new tasks. We investigate the impact of temporal consistency in latent space on task representation learning, showing that enforcing multi-step predictions in latent space encourages task representations that are able to capture task-dependent dynamics while preventing representation collapse. We provide theoretical analysis characterizing sources of error in value estimation and show through extensive experiments on MuJoCo, Contextual DeepMind Control, and MetaWorld benchmarks that temporal consistency significantly improves both zero-shot and few-shot generalization.

Mohammadreza Nakheai, Aidan Scannell, Kevin Luck +1

Aalto University · Toyota Research Institute · Vrije Universiteit Amsterdam

Representation Learning Offline RL Representation Learning for RL Meta-Reinforcement Learning Generalization in RL Temporal Consistency+2

Mar 3, 2026

LLMs for High-Frequency Decision-Making: Normalized Action Reward-Guided Consistency Policy Optimization

While Large Language Models (LLMs) form the cornerstone of sequential decision-making agent development, they have inherent limitations in high-frequency decision tasks. Existing research mainly focuses on discrete embodied decision scenarios with low-frequency and significant semantic differences in state space (e.g., household planning). These methods suffer from limited performance in high-frequency decision-making tasks, since high-precision numerical state information in such tasks undergoes frequent updates with minimal fluctuations, and exhibiting policy misalignment between the learned sub-tasks and composite tasks. To address these issues, this paper proposes Normalized Action Reward guided Consistency Policy Optimization (NAR-CP). 1) Our method first acquires predefined dense rewards from environmental feedback of candidate actions via reward functions, then completes reward shaping through normalization, and theoretically verifies action reward normalization does not impair optimal policy. 2) To reduce policy misalignment in composite tasks, we use LLMs to infer sub-observation candidate actions and generate joint policies, with consistency loss ensuring precise alignment between global semantic policies and sub-semantic policies. Experiments on UAV pursuit, a typical high-frequency task, show our method delivers superior performance on independent and composite tasks with excellent generalization to unseen tasks.

Yang Zhao, Zihao Li, Zhiyu Jiang +3

School of Artificial Intelligence, OPtics and ElectroNics (iOPEN), Northwestern Polytechnical University, Xi’an 710072, China

Policy Learning LLM-Guided RL Reward Shaping

Mar 2, 2026

Learning Therapist Policy from Therapist-Exoskeleton-Patient Interaction

Post-stroke rehabilitation is often necessary for patients to regain proper walking gait. However, the typical therapy process can be exhausting and physically demanding for therapists, potentially reducing therapy intensity, duration, and consistency over time. We propose a Patient-Therapist Force Field (PTFF) to visualize therapist responses to patient kinematics and a Synthetic Therapist (ST) machine learning model to support the therapist in dyadic robot-mediated physical interaction therapy. The first encodes patient and therapist stride kinematics into a shared low-dimensional latent manifold using a Variational Autoencoder (VAE) and models their interaction through a Gaussian Mixture Model (GMM), which learns a probabilistic vector field mapping patient latent states to therapist responses. This representation visualizes patient-therapist interaction dynamics to inform therapy strategies and robot controller design. The latter is implemented as a Long Short-Term Memory (LSTM) network trained on patient-therapist interaction data to predict therapist-applied joint torques from patient kinematics. Trained and validated using leave-one-out cross-validation across eight post-stroke patients, the model was integrated into a ROS-based exoskeleton controller to generate real-time torque assistance based on predicted therapist responses. Offline results and preliminary testing indicate the potential of their use as an alternative approach to post-stroke exoskeleton therapy. The PTFF provides understanding of the therapist's actions while the ST frees the human therapist from the exoskeleton, allowing them to continuously monitor the patient's nuanced condition.

Grayson Snyder, Lorenzo Vianello, Levi Hargrove +2

Center for Robotics and Biosystems, Northwestern University, Evanston, IL, USA. · Legs and Walking Lab, Shirley Ryan AbilityLab, Chicago, IL, USA.

Assistive Robotics Human-Robot Interaction Policy Learning Robot Imitation Learning Ankle Exoskeleton Control

Mar 1, 2026

Evaluating GFlowNet from partial episodes for stable and flexible policy-based training

Generative Flow Networks (GFlowNets) were developed to learn policies for efficiently sampling combinatorial candidates by interpreting their generative processes as trajectories in directed acyclic graphs. In the value-based training workflow, the objective is to enforce the balance over partial episodes between the flows of the learned policy and the estimated flows of the desired policy, implicitly encouraging policy divergence minimization. The policy-based strategy alternates between estimating the policy divergence and updating the policy, but reliable estimation of the divergence under directed acyclic graphs remains a major challenge. This work bridges the two perspectives by showing that flow balance also yields a principled policy evaluator that measures the divergence, and an evaluation balance objective over partial episodes is proposed for learning the evaluator. As demonstrated on both synthetic and real-world tasks, evaluation balance not only strengthens the reliability of policy-based training but also broadens its flexibility by seamlessly supporting parameterized backward policies and enabling the integration of offline data-collection techniques.

Puhua Niu, Shili Wu, Xiaoning Qian

Texas A&M University, College Station, TX 77843, USA · Brookhaven National Laboratory, Upton, NY 11973, USA

Generative Flow Networks Flow-Based Policy Learning Policy Learning

Feb 27, 2026

Pessimistic Auxiliary Policy for Offline Reinforcement Learning

Offline reinforcement learning aims to learn an agent from pre-collected datasets, avoiding unsafe and inefficient real-time interaction. However, inevitable access to out-ofdistribution actions during the learning process introduces approximation errors, causing the error accumulation and considerable overestimation. In this paper, we construct a new pessimistic auxiliary policy for sampling reliable actions. Specifically, we develop a pessimistic auxiliary strategy by maximizing the lower confidence bound of the Q-function. The pessimistic auxiliary strategy exhibits a relatively high value and low uncertainty in the vicinity of the learned policy, avoiding the learned policy sampling high-value actions with potentially high errors during the learning process. Less approximation error introduced by sampled action from pessimistic auxiliary strategy leads to the alleviation of error accumulation. Extensive experiments on offline reinforcement learning benchmarks reveal that utilizing the pessimistic auxiliary strategy can effectively improve the efficacy of other offline RL approaches.

Fan Zhang, Baoru Huang, Xin Zhang

Department of Mathematics, The Hong Kong University of Science and Technology, Hong Kong, China. · University of Liverpool, UK · Department of Mechanical and Aerospace Engineering, The Hong Kong University of Science and Technology, Hong Kong, China.

Offline RL Policy Learning

Feb 27, 2026

General Bayesian Policy Learning

This study proposes a General Bayes framework for policy learning. We consider decision problems in which a decision-maker chooses an action from a given set to maximize expected welfare. Typical examples include treatment choice and portfolio optimization. In such problems, the statistical target is a decision rule, and predicting each potential outcome is not necessarily of primary interest. We formulate this policy-learning problem through loss-based Bayesian updating. Our main technical device is a squared-loss surrogate for welfare maximization. We show that maximizing empirical welfare over a policy class with a quadratic penalty controlled by a tuning parameter

ζ>0

is equivalent to minimizing a scaled squared error in the outcome difference. The resulting General Bayes posterior over decision rules admits two equivalent characterizations: a Gaussian pseudo-likelihood representation and a decision-theoretic loss-based characterization. As one implementation, we introduce GBPLNet, a neural network implementation with a tanh-squashed output. Finally, we establish PAC-Bayes-type guarantees for the surrogate risk and the corresponding penalized welfare.

Masahiro Kato

1Mizuho-DL Financial Technology Co., Ltd., Tokyo, Japan

Generalized Bayesian Inference Policy Learning Bayesian Decision Theory Bayesian Inference

Feb 26, 2026

Multi-agent imitation learning with function approximation: Linear Markov games and beyond

In this work, we present the first theoretical analysis of multi-agent imitation learning (MAIL) in linear Markov games where both the transition dynamics and each agent's reward function are linear in some given features. We demonstrate that by leveraging this structure, it is possible to replace the state-action level "all policy deviation concentrability coefficient" (Freihaut et al., arXiv:2510.09325) with a concentrability coefficient defined at the feature level which can be much smaller than the state-action analog when the features are informative about states' similarity. Furthermore, to circumvent the need for any concentrability coefficient, we turn to the interactive setting. We provide the first, computationally efficient, interactive MAIL algorithm for linear Markov games and show that its sample complexity depends only on the dimension of the feature map

d

. Building on these theoretical findings, we propose a deep MAIL interactive algorithm which clearly outperforms BC on games such as Tic-Tac-Toe and Connect4.

Luca Viano, Till Freihaut, Emanuele Nevali +3

1EPFL · University of Zurich · 3Earth Species Project

Game Theory Imitation Learning Interactive Imitation Learning Policy Learning Linear Function Approximation