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.
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.
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.
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.
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.
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.
At the core of reinforcement learning is the idea of learning beyond the performance in the data. However, scaling such systems has proven notoriously tricky. In contrast, techniques from generative modeling have proven remarkably scalable and are simple to train. In this work, we combine these strengths, by deriving a direct relation between policy improvement and guidance of diffusion models. The resulting framework, CFGRL, is trained with the simplicity of supervised learning, yet can further improve on the policies in the data. On offline RL tasks, we observe a reliable trend -- increased guidance weighting leads to increased performance. Of particular importance, CFGRL can operate without explicitly learning a value function, allowing us to generalize simple supervised methods (e.g., goal-conditioned behavioral cloning) to further prioritize optimality, gaining performance for "free" across the board.
Offline reinforcement learning (RL) aims to learn an effective policy from a static dataset. To alleviate extrapolation errors, existing studies often uniformly regularize the value function or policy updates across all states. However, due to substantial variations in data quality, the fixed regularization strength often leads to a dilemma: Weak regularization strength fails to address extrapolation errors and value overestimation, while strong regularization strength shifts policy learning toward behavior cloning, impeding potential performance enabled by Bellman updates. To address this issue, we propose the selective state-adaptive regularization method for offline RL. Specifically, we introduce state-adaptive regularization coefficients to trust state-level Bellman-driven results, while selectively applying regularization on high-quality actions, aiming to avoid performance degradation caused by tight constraints on low-quality actions. By establishing a connection between the representative value regularization method, CQL, and explicit policy constraint methods, we effectively extend selective state-adaptive regularization to these two mainstream offline RL approaches. Extensive experiments demonstrate that the proposed method significantly outperforms the state-of-the-art approaches in both offline and offline-to-online settings on the D4RL benchmark.
Offline reinforcement learning (RL) has gained traction as a powerful paradigm for learning control policies from pre-collected data, eliminating the need for costly or risky online interactions. While many open-source libraries offer robust implementations of offline RL algorithms, they all rely on datasets composed of experience tuples consisting of state, action, next state, and reward. Managing, curating, and distributing such datasets requires suitable infrastructure. Although static datasets exist for established benchmark problems, no standardized or scalable solution supports developing and sharing datasets for novel or user-defined benchmarks. To address this gap, we introduce PyTupli, a Python-based tool to streamline the creation, storage, and dissemination of benchmark environments and their corresponding tuple datasets. PyTupli includes a lightweight client library with defined interfaces for uploading and retrieving benchmarks and data. It supports fine-grained filtering at both the episode and tuple level, allowing researchers to curate high-quality, task-specific datasets. A containerized server component enables production-ready deployment with authentication, access control, and automated certificate provisioning for secure use. By addressing key barriers in dataset infrastructure, PyTupli facilitates more collaborative, reproducible, and scalable offline RL research.
LLM-as-a-Judge refers to the automatic modeling of preferences for responses generated by Large Language Models (LLMs), which is of significant importance for both LLM evaluation and reward modeling. Although generative LLMs have made substantial progress in various tasks, their performance as LLM-Judge still falls short of expectations. In this work, we propose Think-J, which improves generative LLM-as-a-Judge by learning how to think. We first utilized a small amount of curated data to develop the model with initial judgment thinking capabilities. Subsequently, we optimize the judgment thinking traces based on reinforcement learning (RL). We propose two methods for judgment thinking optimization, based on offline and online RL, respectively. The offline RL requires training a critic model to construct positive and negative examples for learning. The online method defines rule-based reward as feedback for optimization. Experimental results showed that our approach can significantly enhance the evaluation capability of generative LLM-Judge, surpassing both generative and classifier-based LLM-Judge without requiring extra human annotations.
A central challenge in reinforcement learning (RL) is its dependence on extensive real-world interaction data to learn task-specific policies. While recent work demonstrates that large language models (LLMs) can mitigate this limitation by generating synthetic experience (noted as imaginary rollouts) for mastering novel tasks, progress in this emerging field is hindered due to the lack of a standard benchmark. To bridge this gap, we introduce ImagineBench, the first comprehensive benchmark for evaluating offline RL algorithms that leverage both real rollouts and LLM-imaginary rollouts. The key features of ImagineBench include: (1) datasets comprising environment-collected and LLM-imaginary rollouts; (2) diverse domains of environments covering locomotion, robotic manipulation, and navigation tasks; and (3) natural language task instructions with varying complexity levels to facilitate language-conditioned policy learning. Through systematic evaluation of state-of-the-art offline RL algorithms, we observe that simply applying existing offline RL algorithms leads to suboptimal performance on unseen tasks, achieving 35.44% success rate in hard tasks in contrast to 64.37% of method training on real rollouts for hard tasks. This result highlights the need for algorithm advancements to better leverage LLM-imaginary rollouts. Additionally, we identify key opportunities for future research: including better utilization of imaginary rollouts, fast online adaptation and continual learning, and extension to multi-modal tasks. Our code is publicly available at https://github.com/LAMDA-RL/ImagineBench.
Progress in offline reinforcement learning (RL) has been impeded by ambiguous problem definitions and entangled algorithmic designs, resulting in inconsistent implementations, insufficient ablations, and unfair evaluations. Although offline RL explicitly avoids environment interaction, prior methods frequently employ extensive, undocumented online evaluation for hyperparameter tuning, complicating method comparisons. Moreover, existing reference implementations differ significantly in boilerplate code, obscuring their core algorithmic contributions. We address these challenges by first introducing a rigorous taxonomy and a transparent evaluation protocol that explicitly quantifies online tuning budgets. To resolve opaque algorithmic design, we provide clean, minimalistic, single-file implementations of various model-free and model-based offline RL methods, significantly enhancing clarity and achieving substantial speed-ups. Leveraging these streamlined implementations, we propose Unifloral, a unified algorithm that encapsulates diverse prior approaches within a single, comprehensive hyperparameter space, enabling algorithm development in a shared hyperparameter space. Using Unifloral with our rigorous evaluation protocol, we develop two novel algorithms - TD3-AWR (model-free) and MoBRAC (model-based) - which substantially outperform established baselines. Our implementation is publicly available at https://github.com/EmptyJackson/unifloral.
Offline reinforcement learning (RL) aims to learn from historical data without requiring (costly) access to the environment. To facilitate offline RL research, we previously introduced NeoRL, which highlighted that datasets from real-world tasks are often conservative and limited. With years of experience applying offline RL to various domains, we have identified additional real-world challenges. These include extremely conservative data distributions produced by deployed control systems, delayed action effects caused by high-latency transitions, external factors arising from the uncontrollable variance of transitions, and global safety constraints that are difficult to evaluate during the decision-making process. These challenges are underrepresented in previous benchmarks but frequently occur in real-world tasks. To address this, we constructed the extended Near Real-World Offline RL Benchmark (NeoRL-2), which consists of 7 datasets from 7 simulated tasks along with their corresponding evaluation simulators. Benchmarking results from state-of-the-art offline RL approaches demonstrate that current methods often struggle to outperform the data-collection behavior policy, highlighting the need for more effective methods. We hope NeoRL-2 will accelerate the development of reinforcement learning algorithms for real-world applications. The benchmark project page is available at https://github.com/polixir/NeoRL2.
Offline reinforcement learning (RL) aims to optimize a policy by using pre-collected datasets, to maximize cumulative rewards. However, offline reinforcement learning suffers challenges due to the distributional shift between the learned and behavior policies, leading to errors when computing Q-values for out-of-distribution (OOD) actions. To mitigate this issue, policy constraint methods aim to constrain the learned policy's distribution with the distribution of the behavior policy or confine action selection within the support of the behavior policy. However, current policy constraint methods tend to exhibit excessive conservatism, hindering the policy from further surpassing the behavior policy's performance. In this work, we present Only Support Constraint (OSC) which is derived from maximizing the total probability of learned policy in the support of behavior policy, to address the conservatism of policy constraint. OSC presents a regularization term that only restricts policies to the support without imposing extra constraints on actions within the support. Additionally, to fully harness the performance of the new policy constraints, OSC utilizes a diffusion model to effectively characterize the support of behavior policies. Experimental evaluations across a variety of offline RL benchmarks demonstrate that OSC significantly enhances performance, alleviating the challenges associated with distributional shifts and mitigating conservatism of policy constraints. Code is available at https://github.com/MoreanP/OSC.
Sequential decision-making in high-dimensional continuous action spaces, particularly in stochastic environments, faces significant computational challenges. We explore this challenge in the traditional offline RL setting, where an agent must learn how to make decisions based on data collected through a stochastic behavior policy. We present Latent Macro Action Planner (L-MAP), which addresses this challenge by learning a set of temporally extended macro-actions through a state-conditional Vector Quantized Variational Autoencoder (VQ-VAE), effectively reducing action dimensionality. L-MAP employs a (separate) learned prior model that acts as a latent transition model and allows efficient sampling of plausible actions. During planning, our approach accounts for stochasticity in both the environment and the behavior policy by using Monte Carlo tree search (MCTS). In offline RL settings, including stochastic continuous control tasks, L-MAP efficiently searches over discrete latent actions to yield high expected returns. Empirical results demonstrate that L-MAP maintains low decision latency despite increased action dimensionality. Notably, across tasks ranging from continuous control with inherently stochastic dynamics to high-dimensional robotic hand manipulation, L-MAP significantly outperforms existing model-based methods and performs on-par with strong model-free actor-critic baselines, highlighting the effectiveness of the proposed approach in planning in complex and stochastic environments with high-dimensional action spaces.
Online reinforcement learning (RL) enhances policies through direct interactions with the environment, but faces challenges related to sample efficiency. In contrast, offline RL leverages extensive pre-collected data to learn policies, but often produces suboptimal results due to limited data coverage. Recent efforts have sought to integrate offline and online RL in order to harness the advantages of both approaches. However, effectively combining online and offline RL remains challenging due to issues that include catastrophic forgetting, lack of robustness and sample efficiency. In an effort to address these challenges, we introduce A3 RL , a novel method that actively selects data from combined online and offline sources to optimize policy improvement. We provide theoretical guarantee that validates the effectiveness our active sampling strategy and conduct thorough empirical experiments showing that our method outperforms existing state-of-the-art online RL techniques that utilize offline data. Our code will be publicly available at: https://github.com/xuefeng-cs/A3RL.
We present flow Q-learning (FQL), a simple and performant offline reinforcement learning (RL) method that leverages an expressive flow-matching policy to model arbitrarily complex action distributions in data. Training a flow policy with RL is a tricky problem, due to the iterative nature of the action generation process. We address this challenge by training an expressive one-step policy with RL, rather than directly guiding an iterative flow policy to maximize values. This way, we can completely avoid unstable recursive backpropagation, eliminate costly iterative action generation at test time, yet still mostly maintain expressivity. We experimentally show that FQL leads to strong performance across 73 challenging state- and pixel-based OGBench and D4RL tasks in offline RL and offline-to-online RL. Project page: https://seohong.me/projects/fql/
Reinforcement Learning (RL) methods used for solving real-world optimization problems often involve dynamic state-action spaces, larger scale, and sparse rewards, leading to significant challenges in convergence, scalability, and efficient exploration of the solution space. This study introduces GNN-DT, a novel Decision Transformer (DT) architecture that integrates Graph Neural Network (GNN) embedders with a novel residual connection between input and output tokens crucial for handling dynamic environments. By learning from previously collected trajectories, GNN-DT tackles the sparse rewards limitations of online RL algorithms and delivers high-quality solutions in real-time. We evaluate GNN-DT on the complex electric vehicle (EV) charging optimization problem and prove that its performance is superior and requires significantly fewer training trajectories, thus improving sample efficiency compared to existing DT and offline RL baselines. Furthermore, GNN-DT exhibits robust generalization to unseen environments and larger action spaces, addressing a critical gap in prior offline and online RL approaches.
Sparse continuous policies are distributions that can choose some actions at random yet keep strictly zero probability for the other actions, which are radically different from the Gaussian. They have important real-world implications, e.g. in modeling safety-critical tasks like medicine. The combination of offline reinforcement learning and sparse policies provides a novel paradigm that enables learning completely from logged datasets a safety-aware sparse policy. However, sparse policies can cause difficulty with the existing offline algorithms which require evaluating actions that fall outside of the current support. In this paper, we propose the first offline policy optimization algorithm that tackles this challenge: Fat-to-Thin Policy Optimization (FtTPO). Specifically, we maintain a fat (heavy-tailed) proposal policy that effectively learns from the dataset and injects knowledge to a thin (sparse) policy, which is responsible for interacting with the environment. We instantiate FtTPO with the general $q$-Gaussian family that encompasses both heavy-tailed and sparse policies and verify that it performs favorably in a safety-critical treatment simulation and the standard MuJoCo suite. Our code is available at \url{https://github.com/lingweizhu/fat2thin}.