Offline RL (ORL) promises safe and sample-efficient deployment but existing methods rely on undocumented online interactions for hyperparameter tuning and lack reliable fully offline estimates of initial online performance. We introduce SOReL, a fully offline Bayesian model-based RL method that learns a posterior over dynamics, estimates policy value via predictive uncertainty, and enables complete offline hyperparameter selection. We further propose TOReL, which extends this tuning framework to arbitrary model-free and model-based ORL algorithms. We provide a regret analysis showing that Bayesian offline RL achieves the minimax-optimal parametric rate under standard regularity conditions. Together, our methods establish a practical and theoretically grounded framework for fully offline RL.
Offline Reinforcement Learning (RL) aims to learn effective policies from a static dataset without requiring further agent-environment interactions. However, its practical adoption is often hindered by the need for explicit reward annotations, which can be costly to engineer or difficult to obtain retrospectively. To address this, we propose ReLOAD (Reinforcement Learning with Offline Reward Annotation via Distillation), a novel reward annotation framework for offline RL. Unlike existing methods that depend on complex alignment procedures, our approach adapts Random Network Distillation (RND) to generate intrinsic rewards from expert demonstrations using a simple yet effective embedding discrepancy measure. First, we train a predictor network to mimic a fixed target network's embeddings based on expert state transitions. Later, the prediction error between these networks serves as a reward signal for each transition in the static dataset. This mechanism provides a structured reward signal without requiring handcrafted reward annotations. We provide a formal theoretical construct that offers insights into how RND prediction errors effectively serve as intrinsic rewards by distinguishing expert-like transitions. Experiments on the D4RL benchmark demonstrate that ReLOAD enables robust offline policy learning and achieves performance competitive with traditional reward-annotated methods.
We present Step-wise Policy for Rare-tool Knowledge (SPaRK), a novel reinforcement learning framework that teaches large language models to explore diverse tool usage patterns beyond conventional high-temperature sampling. Building on recent advances in step-wise reinforcement learning, we introduce a dual-objective reward system that simultaneously optimizes for answer quality and tool diversity, training a Llama-3.1 8B model through offline PPO on synthetically generated trajectories from the MMLU-Pro dataset. Our approach uniquely employs a rarity-first exploitation strategy where a GPT-4o judge scores candidate actions across eight distinct tools plus chain-of-thought reasoning, with the policy favoring less-frequently used but still viable tools to encourage systematic exploration. Empirical results demonstrate that SPaRK achieves competitive performance across 14 MMLU-Pro categories while exhibiting significantly higher entropy in tool selection compared to both baseline and supervised fine-tuning approaches, suggesting that algorithmic exploration through explicit tool diversity can enhance reasoning capabilities without sacrificing accuracy.
Accurate bandwidth estimation (BWE) is critical for real-time communication (RTC) systems. Traditional heuristic approaches offer limited adaptability under dynamic networks, while online reinforcement learning (RL) suffers from high exploration costs and potential service disruptions. Offline RL, which leverages high-quality data collected from real-world environments, offers a promising alternative. However, challenges such as out-of-distribution (OOD) actions, policy extraction from behaviorally diverse datasets, and reliable deployment in production systems remain unsolved. We propose RBWE, a robust bandwidth estimation framework based on offline RL that integrates Q-ensemble (an ensemble of Q-functions) with a Gaussian mixture policy to mitigate OOD risks and enhance policy learning. A fallback mechanism ensures deployment stability by switching to heuristic methods under high uncertainty. Experimental results show that RBWE reduces overestimation errors by 18% and improves the 10th percentile Quality of Experience (QoE) by 18.6%, demonstrating its practical effectiveness in real-world RTC applications.
Offline reinforcement learning (offline RL) algorithms often require additional constraints or penalty terms to address distribution shift issues, such as adding implicit or explicit policy constraints during policy optimization to reduce the estimation bias of functions. This paper focuses on a limitation of the Advantage-Weighted Regression family (AWRs), i.e., the potential for learning over-conservative policies due to data corruption, specifically the poor explorations in suboptimal offline data. We study it from two perspectives: (1) how poor explorations impact the theoretically optimal policy based on KL divergence, and (2) how such poor explorations affect the approximation of the theoretically optimal policy. We prove that such over-conservatism is mainly caused by the sensitivity of the loss function for policy optimization to poor explorations, and the proportion of poor explorations in offline datasets. To address this concern, we propose Corruption-Averse Advantage-Weighted Regression (CAWR), which incorporates a set of robust loss functions during policy optimization and an advantage-based prioritized experience replay method to filter out poor explorations. Numerical experiments on the D4RL benchmark show that our method can learn superior policies from suboptimal offline data, significantly enhancing the performance of policy optimization.
Generative models such as diffusion and flow-matching offer expressive policies for offline reinforcement learning (RL) by capturing rich, multimodal action distributions, but their iterative sampling introduces high inference costs and training instability due to gradient propagation across sampling steps. We propose the \textit{Single-Step Completion Policy} (SSCP), a generative policy trained with an augmented flow-matching objective to predict direct completion vectors from intermediate flow samples, enabling accurate, one-shot action generation. In an off-policy actor-critic framework, SSCP combines the expressiveness of generative models with the training and inference efficiency of unimodal policies, without requiring long backpropagation chains. Our method scales effectively to offline, offline-to-online, and online RL settings, offering substantial gains in speed and adaptability over diffusion-based baselines. We further extend SSCP to goal-conditioned RL, enabling flat policies to exploit subgoal structures without explicit hierarchical inference. SSCP achieves strong results across standard offline RL and behavior cloning benchmarks, positioning it as a versatile, expressive, and efficient framework for deep RL and sequential decision-making.
We study offline reinforcement learning in average-reward MDPs, which presents increased challenges from the perspectives of distribution shift and non-uniform coverage, and has been relatively underexamined from a theoretical perspective. While previous work obtains performance guarantees under single-policy data coverage assumptions, such guarantees utilize additional complexity measures which are uniform over all policies, such as the uniform mixing time. We develop sharp guarantees depending only on the target policy, specifically the bias span and a novel policy hitting radius, yielding the first fully single-policy sample complexity bound for average-reward offline RL. We are also the first to handle general weakly communicating MDPs, contrasting restrictive structural assumptions made in prior work. To achieve this, we introduce an algorithm based on pessimistic discounted value iteration enhanced by a novel quantile clipping technique, which enables the use of a sharper empirical-span-based penalty function. Our algorithm also does not require any prior parameter knowledge for its implementation. Remarkably, we show via hard examples that learning under our conditions requires coverage assumptions beyond the stationary distribution of the target policy, distinguishing single-policy complexity measures from previously examined cases. We also develop lower bounds nearly matching our main result.
In this paper, we investigate the use of small datasets in the context of offline reinforcement learning (RL). While many common offline RL benchmarks employ datasets with over a million data points, many offline RL applications rely on considerably smaller datasets. We show that offline RL algorithms can overfit on small datasets, resulting in poor performance. To address this challenge, we introduce "Sparse-Reg": a regularization technique based on sparsity to mitigate overfitting in offline reinforcement learning, enabling effective learning in limited data settings and outperforming state-of-the-art baselines in continuous control.
Question answering (QA) agents automatically answer questions posed in natural language. In this work, we learn to ask clarifying questions in QA agents. The key idea in our method is to simulate conversations that contain clarifying questions and learn from them using reinforcement learning (RL). To make RL practical, we propose and analyze offline RL objectives that can be viewed as reward-weighted supervised fine-tuning (SFT) and easily optimized in large language models. Our work stands in a stark contrast to recently proposed methods, based on SFT and direct preference optimization, which have additional hyper-parameters and do not directly optimize rewards. We compare to these methods empirically and report gains in both optimized rewards and language quality.
Invasive mechanical ventilation (MV) is a life-sustaining therapy for critically ill patients in the intensive care unit (ICU). However, optimizing its settings remains a complex and error-prone process due to patient-specific variability. While Offline Reinforcement Learning (RL) shows promise for MV control, current stateof-the-art (SOTA) methods struggle with the hybrid (continuous and discrete) nature of MV actions. Discretizing the action space limits available actions due to exponential growth in combinations and introduces distribution shifts that can compromise safety. In this paper, we propose optimizations that build upon prior work in action space reduction to address the challenges of discrete action spaces. We also adapt SOTA offline RL algorithms (IQL and EDAC) to operate directly on hybrid action spaces, thereby avoiding the pitfalls of discretization. Additionally, we introduce a clinically grounded reward function based on ventilator-free days and physiological targets, which provides a more meaningful optimization objective compared to traditional sparse mortality-based rewards. Our findings demonstrate that AI-assisted MV optimization may enhance patient safety and enable individualized lung support, representing a significant advancement toward intelligent, data-driven critical care solutions.
Elastic Decision Transformers (EDTs) have proved to be particularly successful in offline reinforcement learning, offering a flexible framework that unifies sequence modeling with decision-making under uncertainty. Recent research has shown that incorporating intrinsic motivation mechanisms into EDTs improves performance across exploration tasks, yet the representational mechanisms underlying these improvements remain unexplored. In this paper, we introduce a systematic post-hoc explainability framework to analyze how intrinsic motivation shapes learned embeddings in EDTs. Through statistical analysis of embedding properties (including covariance structure, vector magnitudes, and orthogonality), we reveal that different intrinsic motivation variants create fundamentally different representational structures. Our analysis demonstrates environment-specific correlation patterns between embedding metrics and performance that explain why intrinsic motivation improves policy learning. These findings show that intrinsic motivation operates beyond simple exploration bonuses, acting as a representational prior that shapes embedding geometry in biologically plausible ways, creating environment-specific organizational structures that facilitate better decision-making.
Deep reinforcement learning (RL) has achieved significant success, yet its application in real-world scenarios is often hindered by a lack of robustness to environmental uncertainties. To solve this challenge, some robust RL algorithms have been proposed, but most are limited to tabular settings. In this work, we propose Distributionally Robust Soft Actor-Critic (DR-SAC), a novel algorithm designed to enhance the robustness of the state-of-the-art Soft Actor-Critic (SAC) algorithm. DR-SAC aims to maximize the expected value with entropy against the worst possible transition model lying in an uncertainty set. A distributionally robust version of the soft policy iteration is derived with a convergence guarantee. For settings where nominal distributions are unknown, such as offline RL, a generative modeling approach is proposed to estimate the required nominal distributions from data. Furthermore, experimental results on a range of continuous control benchmark tasks demonstrate our algorithm achieves up to $9.8$ times the average reward of the SAC baseline under common perturbations. Additionally, compared with existing robust reinforcement learning algorithms, DR-SAC significantly improves computing efficiency and applicability to large-scale problems.
Sample efficiency and exploration remain critical challenges in Deep Reinforcement Learning (DRL), particularly in complex domains. Offline RL, which enables agents to learn optimal policies from static, pre-collected datasets, has emerged as a promising alternative. However, offline RL is constrained by issues such as out-of-distribution (OOD) actions that limit policy performance and generalization. To overcome these limitations, we propose Meta Offline-Online Reinforcement Learning (MOORL), a hybrid framework that unifies offline and online RL for efficient and scalable learning. While previous hybrid methods rely on extensive design components and added computational complexity to utilize offline data effectively, MOORL introduces a meta-policy that seamlessly adapts across offline and online trajectories. This enables the agent to leverage offline data for robust initialization while utilizing online interactions to drive efficient exploration. Our theoretical analysis demonstrates that the hybrid approach enhances exploration by effectively combining the complementary strengths of offline and online data. Furthermore, we demonstrate that MOORL learns a stable Q-function without added complexity. Extensive experiments on 28 tasks from the D4RL and V-D4RL benchmarks validate its effectiveness, showing consistent improvements over state-of-the-art offline and hybrid RL baselines. With minimal computational overhead, MOORL achieves strong performance, underscoring its potential for practical applications in real-world scenarios.
We introduce a novel task of clustering trajectories from offline reinforcement learning (RL) datasets, where each cluster center represents the policy that generated its trajectories. By leveraging the connection between the KL-divergence of offline trajectory distributions and a mixture of policy-induced distributions, we formulate a natural clustering objective. To solve this, we propose Policy-Guided K-means (PG-Kmeans) and Centroid-Attracted Autoencoder (CAAE). PG-Kmeans iteratively trains behavior cloning (BC) policies and assigns trajectories based on policy generation probabilities, while CAAE resembles the VQ-VAE framework by guiding the latent representations of trajectories toward the vicinity of specific codebook entries to achieve clustering. Theoretically, we prove the finite-step convergence of PG-Kmeans and identify a key challenge in offline trajectory clustering: the inherent ambiguity of optimal solutions due to policy-induced conflicts, which can result in multiple equally valid but structurally distinct clusterings. Experimentally, we validate our methods on the widely used D4RL dataset and custom GridWorld environments. Our results show that both PG-Kmeans and CAAE effectively partition trajectories into meaningful clusters. They offer a promising framework for policy-based trajectory clustering, with broad applications in offline RL and beyond.
Although diffusion models have achieved strong results in decision-making tasks, their slow inference speed remains a key limitation. While the consistency model offers a potential solution, its applications to decision-making often struggle with suboptimal demonstrations or rely on complex concurrent training of multiple networks. In this work, we propose a novel approach to consistency distillation for offline reinforcement learning that directly incorporates reward optimization into the distillation process. Our method enables single-step generation while maintaining higher performance and simpler training. Empirical evaluations on the Gym MuJoCo benchmarks and long horizon planning demonstrate that our approach can achieve an 8.7% improvement over previous state-of-the-art while offering up to 142x speedup over diffusion counterparts in inference time.
Stationary Distribution Correction Estimation (DICE) addresses the mismatch between the stationary distribution induced by a policy and the target distribution required for reliable off-policy evaluation (OPE) and policy optimization. DICE-based offline constrained RL particularly benefits from the flexibility of DICE, as it simultaneously maximizes return while estimating costs in offline settings. However, we have observed that recent approaches designed to enhance the offline RL performance of the DICE framework inadvertently undermine its ability to perform OPE, making them unsuitable for constrained RL scenarios. In this paper, we identify the root cause of this limitation: their reliance on a semi-gradient optimization, which solves a fundamentally different optimization problem and results in failures in cost estimation. Building on these insights, we propose a novel method to enable OPE and constrained RL through semi-gradient DICE. Our method ensures accurate cost estimation and achieves state-of-the-art performance on the offline constrained RL benchmark, DSRL.
In sequential decision-making problems, Return-Conditioned Supervised Learning (RCSL) has gained increasing recognition for its simplicity and stability in modern decision-making tasks. Unlike traditional offline reinforcement learning (RL) algorithms, RCSL frames policy learning as a supervised learning problem by taking both the state and return as input. This approach eliminates the instability often associated with temporal difference (TD) learning in offline RL. However, RCSL has been criticized for lacking the stitching property, meaning its performance is inherently limited by the quality of the policy used to generate the offline dataset. To address this limitation, we propose a principled and simple framework called Reinforced RCSL. The key innovation of our framework is the introduction of a concept we call the in-distribution optimal return-to-go. This mechanism leverages our policy to identify the best achievable in-dataset future return based on the current state, avoiding the need for complex return augmentation techniques. Our theoretical analysis demonstrates that Reinforced RCSL can consistently outperform the standard RCSL approach. Empirical results further validate our claims, showing significant performance improvements across a range of benchmarks.
We study the off-dynamics offline reinforcement learning problem, where the goal is to learn a policy from offline datasets collected from source and target domains with mismatched transition. Existing off-dynamics offline RL methods typically either filter source transitions that resemble those of the target domain or apply reward augmentation to source data, both constrained by the limited transitions available from the target domain. As a result, the learned policy is unable to explore target domain beyond the offline datasets. We propose MOBODY, a Model-Based Off-Dynamics offline RL algorithm that addresses this limitation by enabling exploration of the target domain via learned dynamics. MOBODY generates new synthetic transitions in the target domain through model rollouts, which are used as data augmentation during offline policy learning. Unlike existing model-based methods that learn dynamics from a single domain, MOBODY tackles the challenge of mismatched dynamics by leveraging both source and target datasets. Directly merging these datasets can bias the learned model toward source dynamics. Instead, MOBODY learns target dynamics by discovering a shared latent representation of states and transitions across domains through representation learning. To stabilize training, MOBODY incorporates a behavior cloning loss that regularizes the policy. Specifically, we introduce a Q-weighted behavior cloning loss that regularizes the policy toward actions with high target-domain Q-values, rather than uniformly imitating all actions in the dataset. These Q-values are learned from an enhanced target dataset composed of offline target data, augmented source data, and rollout data from the learned target dynamics. We evaluate MOBODY on MuJoCo benchmarks and show that it significantly outperforms state-of-the-art baselines, with especially pronounced improvements in challenging scenarios.
Offline Reinforcement Learning (RL) struggles with distributional shifts, leading to the $Q$-value overestimation for out-of-distribution (OOD) actions. Existing methods address this issue by imposing constraints; however, they often become overly conservative when evaluating OOD regions, which constrains the $Q$-function generalization. This over-constraint issue results in poor $Q$-value estimation and hinders policy improvement. In this paper, we introduce a novel approach to achieve better $Q$-value estimation by enhancing $Q$-function generalization in OOD regions within Convex Hull and its Neighborhood (CHN). Under the safety generalization guarantees of the CHN, we propose the Smooth Bellman Operator (SBO), which updates OOD $Q$-values by smoothing them with neighboring in-sample $Q$-values. We theoretically show that SBO approximates true $Q$-values for both in-sample and OOD actions within the CHN. Our practical algorithm, Smooth Q-function OOD Generalization (SQOG), empirically alleviates the over-constraint issue, achieving near-accurate $Q$-value estimation. On the D4RL benchmarks, SQOG outperforms existing state-of-the-art methods in both performance and computational efficiency.
Real-world datasets collected from sensors or human inputs are prone to noise and errors, posing significant challenges for applying offline reinforcement learning (RL). While existing methods have made progress in addressing corrupted actions and rewards, they remain insufficient for handling corruption in high-dimensional state spaces and for cases where multiple elements in the dataset are corrupted simultaneously. Diffusion models, known for their strong denoising capabilities, offer a promising direction for this problem-but their tendency to overfit noisy samples limits their direct applicability. To overcome this, we propose Ambient Diffusion-Guided Dataset Recovery (ADG), a novel approach that pioneers the use of diffusion models to tackle data corruption in offline RL. First, we introduce Ambient Denoising Diffusion Probabilistic Models (DDPM) from approximated distributions, which enable learning on partially corrupted datasets with theoretical guarantees. Second, we use the noise-prediction property of Ambient DDPM to distinguish between clean and corrupted data, and then use the clean subset to train a standard DDPM. Third, we employ the trained standard DDPM to refine the previously identified corrupted data, enhancing data quality for subsequent offline RL training. A notable strength of ADG is its versatility-it can be seamlessly integrated with any offline RL algorithm. Experiments on a range of benchmarks, including MuJoCo, Kitchen, and Adroit, demonstrate that ADG effectively mitigates the impact of corrupted data and improves the robustness of offline RL under various noise settings, achieving state-of-the-art results.