论文检索

Theory · Reinforcement Learning and Planning

Alexander Ryabchenko, Wenlong Mou

We introduce Action-Triggered Sporadically Traceable Markov Decision Processes (ATST-MDPs), a novel reinforcement learning framework for partial observability in which full state observations occur stochastically at each step, with probability determined by the chosen action. We derive Bellman equations tailored to this setting and establish the existence of an optimal policy. Exploiting the fact that sporadic observations reveal the full state, we provide an equivalent reformulation in which, upon each state observation, agents commit to a sequence of actions until the next observation. Under the linear MDP assumption, we show that the resulting action sequence value functions admit linear representations in a finite-dimensional feature map, enabling standard regression-based methods. As an application, we derive ST-LSVI-UCB, an optimistic algorithm achieving regret $\widetilde{O}(\sqrt{Kd^3(1-\gamma)^{-3}})$ for episodic learning with geometrically distributed horizons, where $K$ is the number of episodes, $d$ the feature dimension, and $\gamma$ the discount factor (continuation probability), matching the known rate for linear MDPs with full observability.

Deep Learning · Large Language Models

Jiaxi Liu, Yanzuo Jiang, Guibin Zhang, Zihan Zhang, Heng Chang, Zhenfei Yin, Qibing Ren, Junchi Yan

Planning has become a central capability for contemporary agent systems in navigating complex, long-horizon tasks, yet existing approaches predominantly rely on fixed, hand-crafted planning structures that lack the flexibility to adapt to the structural diversity of open-ended problems. To address this limitation, we introduce TodoEvolve, a meta-planning paradigm that autonomously synthesizes and dynamically revises task-specific planning architectures. Specifically, we first construct PlanFactory, a modular design space that standardizes diverse planning paradigms within a unified codebase encompassing topology, initialization, adaptation, and navigation, thereby providing a common interface for heterogeneous planning patterns. Leveraging PlanFactory, we collect high-quality planning trajectories and train Todo-14B via \textit{Impedance-Guided Preference Optimization} (IGPO), a multi-objective reinforcement learning objective that encourages the generation of planning systems that are performant, stable, and token-efficient across arbitrary tasks and agent backbones. Empirical evaluations on five agentic benchmarks demonstrate that TodoEvolve consistently surpasses carefully engineered planning modules while maintaining economical API costs and runtime overhead.

Reinforcement Learning · Deep RL

Raj Ghugare, Michał Bortkiewicz, Alicja Ziarko, Benjamin Eysenbach

How does the amount of compute available to a reinforcement learning (RL) policy affect its learning? Can policies using a fixed amount of parameters, still benefit from additional compute? The standard RL framework does not provide a language to answer these questions formally. Empirically, deep RL policies are often parameterized as neural networks with static architectures, conflating the amount of compute and the number of parameters. In this paper, we formalize compute bounded policies and prove that policies which use more compute can solve problems and generalize to longer-horizon tasks that are outside the scope of policies with less compute. Building on prior work in algorithmic learning and model-free planning, we propose a minimal architecture that can use a variable amount of compute. Our experiments complement our theory. On a set 31 different tasks spanning online and offline RL, we show that $(1)$ this architecture achieves stronger performance simply by using more compute, and $(2)$ stronger generalization on longer-horizon test tasks compared to standard feedforward networks or deep residual network using upto 5 times more parameters.

Reinforcement Learning · Batch/Offline

Jinu Hyeon, Woobin Park, Hongjoon Ahn, Taesup Moon

Hierarchical policies in offline goal-conditioned reinforcement learning (GCRL) addresses long-horizon tasks by decomposing control into high-level subgoal planning and low-level action execution. A critical design choice in such architectures is the goal representation—the compressed encoding of goals that serves as the interface between these levels. Existing approaches commonly derive goal representations while learning value functions, implicitly assuming that preserving information sufficient for value estimation is adequate for optimal control. We show that this assumption can fail, even when the value estimation is exact, as such representations may collapse goal states that need to be differentiated for action learning. To address this, we introduce an information-theoretic framework that defines *action sufficiency*, a condition on goal representations necessary for optimal action selection. We prove that value sufficiency does not imply action sufficiency and empirically verify that the latter is more strongly associated with control success in a discrete environment. We further demonstrate that standard log-loss training of low-level policies naturally induces action-sufficient representations. Our experimental results a popular benchmark demonstrate that our actor-derived representations consistently outperform representations learned via value estimation.

Theory · Reinforcement Learning and Planning

Sreejeet Maity, Aritra Mitra

We study the problem of learning the optimal policy in a discounted, infinite-horizon reinforcement learning (RL) setting in the presence of adversarially corrupted rewards. To address this problem, we develop a novel robust variant of the Q-learning algorithm and analyze it under the challenging asynchronous sampling model with time-correlated data. Despite corruption, we prove that the finite-time guarantees of our approach match existing bounds, up to an additive term that scales with the fraction of corrupted samples. We also establish an information-theoretic lower bound, revealing that our guarantees are near-optimal. Notably, our algorithm is agnostic to the underlying reward distribution and provides the first finite-time robustness guarantees for asynchronous Q-learning. A key element of our analysis is a refined Azuma-Hoeffding inequality for almost-martingales, which may have broader applicability in the study of RL algorithms.

Reinforcement Learning · Multi-agent

Qinhong Zhou, Chuang Gan, Anoop Cherian

Embodied agents operating in decentralized and partially observable environments have attracted growing attention in recent years. However, existing large language model (LLM)–based agents often exhibit behaviors that are misaligned with their partners or inconsistent with the environment state, leading to inefficient cooperation and poor task success. To address this challenge, we propose a novel framework, Learning Laws for Cooperation (LLawCo), that enables embodied agents to autonomously align with both their partners and task objectives. Our framework allows agents to reflect on past failures to extract misaligned behavioral patterns, which are used to derive high-level behavioral laws (e.g., “Talk when necessary”, “Wait for partner”). These laws are explicitly incorporated into the agents’ chains of thought via supervised fine-tuning, aligning their reasoning with task requirements and the behavior of other agents. To evaluate our approach, we introduce PARTNR-Dialog, a large-scale multi-agent communicative and cooperative planning benchmark built on the PARTNR environment. Experiments on existing tasks and our new benchmark demonstrate significant improvements in cooperative efficiency and task success rates. Across four backbone LLMs, our method achieves average success rate improvements of 4.5% on the PARTNR-Dialog benchmark and 6.8% on the TDW-MAT benchmark over state-of-the-art open-source communicative agent frameworks.

Theory · Reinforcement Learning and Planning

David Bossens, Atsushi Nitanda

Safety is an essential requirement for reinforcement learning systems. The newly emerging framework of robust constrained Markov decision processes allows learning policies that satisfy long-term constraints while providing guarantees under epistemic uncertainty. This paper presents mirror descent policy optimisation for robust constrained Markov decision processes, making use of policy gradient techniques to optimise both the policy (as a maximiser) and the transition kernel (as an adversarial minimiser) on the Lagrangian representing a constrained Markov decision process. Our proposed algorithm obtains an $\tilde{\mathcal{O}}\left(1/T^{1/3}\right)$ convergence rate in the sample-based robust constrained Markov decision process setting. The paper also contributes an algorithm for approximate gradient descent in the space of transition kernels, which is of independent interest for designing adversarial environments in general Markov decision processes. Experiments confirm the benefits of mirror descent policy optimisation in constrained and unconstrained optimisation, and significant improvements are observed in robustness tests when compared to baseline policy optimisation algorithms.

Reinforcement Learning · Planning

Jiangweizhi Peng, Yuanxin Liu, Ruida Zhou, Charles Fleming, Zhaoran Wang, Alfredo Garcia, Mingyi Hong

Training LLMs as interactive agents for multi-turn decision-making remains challenging, particularly in long-horizon tasks with sparse and delayed rewards, where agents must execute extended sequences of actions before receiving meaningful feedback. Most existing reinforcement learning (RL) methods model LLM agents as *flat* policies operating at a single time scale, selecting an action at each turn. In sparse-reward settings, this forces the agent to infer long-range dependencies solely from distant end-of-trajectory signals, often leading to inefficient learning and unstable behavior in complex environments. We propose **HiPER**, a novel **Hi**erarchical **P**lan–**E**xecute **R**L framework that jointly models and optimizes high-level subgoal planning and low-level action execution for LLM agents to overcome flat RL's brittle long-horizon behavior and weak credit assignment under sparse outcome feedback. By maintaining persistent subgoals across multiple turns and explicitly deciding when to switch between them, HiPER introduces structured intermediate decision points that facilitate learning under sparse feedback, converting implicit multi-turn structure into learnable decisions at different time scales. To enable effective training, we introduce Hierarchical Advantage Estimation (HAE), a two-timescale policy gradient method that assigns credit to both action execution and subgoal transitions and achieves variance reduction relative to flat advantage estimation. Empirically, HiPER achieves state-of-the-art performance on challenging interactive benchmarks, reaching 97.4\% success on ALFWorld (+6.6\% over the best prior method) and 83.3\% on WebShop, with especially large gains on long-horizon tasks requiring multiple dependent subtasks. These results highlight the importance of explicit hierarchical decomposition for scalable RL training of multi-turn LLM agents.

Theory · Reinforcement Learning and Planning

Yivan Zhang, Ziyan Luo, Manuel Baltieri

State abstraction plays a central role in scaling reinforcement learning to complex but structured systems. In studying such systems, a wide range of behavioral structures have been proposed in reinforcement learning, including value functions, invariants, bisimulation relations, and behavioral metrics. However, a general principle for determining what structures are provably preserved under state abstraction is still lacking. In this paper, we present a unified framework for defining and analyzing behavioral structures in reinforcement learning. Our framework provides a compositional way to specify behavioral semantics and metrics based on local, one-step descriptions of system dynamics. Using this framework, we establish results showing how behavioral structures can be safely transferred between abstract and concrete systems. We further show how to construct quantitative metrics from logical behavioral semantics with soundness guarantees. Together, these results provide a principled foundation for reasoning about behaviors under state abstraction in reinforcement learning and offer reusable definition and proof principles for a broad class of behavioral structures in reinforcement learning.

Social Aspects · Safety

Zhenyu Pan, Yiting Zhang, Zhuo Liu, Yolo Tang, Zeliang Zhang, Haozheng Luo, Chenwei Xu, Yuwei Han, Jianshu Zhang, Dennis Wu 等

LLM-based multi-agent systems excel at planning, tool use, and role coordination, but their openness and interaction complexity also expose them to jailbreak and adversarial collaboration. Existing defenses fall into two lines: (i) self-verification that asks each agent to pre-filter unsafe instructions before execution, and (ii) external guard modules that police behaviors. The former often underperforms because a standalone agent lacks sufficient capacity to detect cross-agent unsafe chains and delegation-induced risks; the latter increases system overhead and creates a single-point-of-failure—once compromised, system-wide safety collapses, and adding more guards worsens cost and complexity. To solve these challenges, we propose AdvEvo-MARL, a co-evolutionary multi-agent reinforcement learning framework that internalizes safety into task agents. Rather than relying on external guards, AdvEvo-MARL jointly optimizes attackers (which synthesize evolving jailbreak prompts) and defenders (task agents trained to both accomplish their duties and resist attacks) in adversarial learning environments. To stabilize learning and foster cooperation, we introduce a public baseline for advantage estimation: agents within the same functional group share a group-level mean-return baseline, enabling lower-variance updates and stronger intra-group coordination. Across representative attack scenarios, AdvEvo-MARL consistently keeps attack-success rate (ASR) below 20\%, whereas baselines reach up to 38.33\%, while preserving or even improving task accuracy (up to +3.67\%). These results show that safety and utility can be jointly improved without relying on extra guard agents or added system overhead.

Theory · Reinforcement Learning and Planning

Jose Aguilar Escamilla, Haoyang Hong, Jiawei Li, Haoyu Zhao, Xuezhou Zhang, Sanghyun Hong, Huazheng Wang

We study reward poisoning attacks in reinforcement learning (RL), where an adversary manipulates rewards within constrained budgets to force the target RL agent to adopt a policy that aligns with the attacker's objectives. Prior works on reward poisoning mainly focused on sufficient conditions to design a successful attacker, while only a few studies discussed the infeasibility of targeted attacks. This paper provides the first precise necessity and sufficiency characterization of the attackability of a linear MDP under reward poisoning attacks. Our characterization draws a bright line between the vulnerable RL instances, and the intrinsically robust ones which cannot be attacked without large costs even running vanilla non-robust RL algorithms. Our theory extends beyond linear MDPs---by approximating deep RL environments as linear MDPs, we show that our theoretical framework effectively distinguishes the attackability and efficiently attacks the vulnerable ones, demonstrating both the theoretical and practical significance of our characterization.

Reinforcement Learning · Multi-agent

Junsung Kim, Ilia Mireskandari, Seungwan Son, Yifan Zhou, Khizer Shahid, Dylan Dai

Autonomous agents for machine learning engineering have advanced rapidly, yet comparing their effectiveness remains difficult. Existing systems combine different techniques---multi-agent decomposition, iterative refinement, memory management, and planning---in varying configurations, making it unclear which components actually drive performance. Complicating evaluation, existing benchmarks rely on historical competitions whose data likely contaminates LLM training corpora and whose static baselines reflect outdated human performance. To address this, we conduct over 4,000 controlled experiments systematically ablating architectural components, alongside K-live, a new benchmark of 25 active competitions that provides a contamination-free, dynamic evaluation environment. Our findings challenge common design assumptions: iterative feedback contributes more than architectural complexity, and multi-agent coordination can hurt as often as it helps. These results provide concrete guidance for practitioners building ML engineering agents.

Reinforcement Learning · Planning

Jonathan Spieler, Sven Behnke

State-of-the-art model-based Reinforcement Learning (RL) approaches either use gradient-free, population-based methods for planning, learned policy networks, or a combination of policy networks and planning. Hybrid approaches that combine Model Predictive Control (MPC) with a learned model and a policy prior to leverage the advantages of both paradigms have shown promising results. However, these approaches typically rely on gradient-free optimization methods, which can be computationally expensive for high-dimensional control tasks. While gradient-based methods are a promising alternative, recent works have empirically shown that gradient-based methods often perform worse than their gradient-free counterparts. We propose Dream-MPC, a novel approach that generates few candidate trajectories from a rolled-out policy and optimizes each trajectory by gradient ascent using a learned world model, uncertainty regularization and amortization of optimization iterations over time by reusing previously optimized actions. Our results on 24 continuous control tasks show that Dream-MPC can significantly improve the performance of the underlying policy and can outperform gradient-free MPC and state-of-the-art baselines. We will open source our code and more at https://dream-mpc.github.io.

Deep Learning · Generative Models and Autoencoders

Daiheng Zhang, Shiyang Zhang, Sizhuang He, Yangtian Zhang, Syed Rizvi, David van Dijk

Discrete biological sequence optimization demands iterative refinement while satisfying strict syntactic constraints. Diffusion-based approaches provide strong progressive refinement but are not naturally aligned with discrete, grammar-constrained edit operations, whereas autoregressive LLMs readily produce valid sequences yet often lack explicit long-horizon planning. To close this gap, we introduce *STRIDE* (Sequence Trajectory Refinement via Internalized Denoising Emulation), a post-training framework that recasts optimization as an intrinsic reasoning problem in edit space. Rather than relying on external agentic search loops, *STRIDE* trains an LLM to emit a full trajectory of atomic edits as explicit Chain-of-Thought, effectively internalizing a trajectory-based refinement policy under discrete constraints. We instantiate *STRIDE* with a curriculum that combines supervised fine-tuning on Levenshtein-aligned shortest-edit demonstrations with GRPO-style reinforcement learning (and variants) to align edit trajectories with task rewards. Across protein and molecule optimization benchmarks, *STRIDE* consistently outperforms a diverse set of baselines, while producing candidates that maintain high structural validity and achieve improved target properties.

Deep Learning · Large Language Models

Shashwat Goel, Rishi Hazra, Dulhan Jayalath, Timon Willi, Parag Jain, Shen, Ilias Leontiadis, Francesco Barbieri, Yoram Bachrach, Jonas Geiping 等

AI co-scientists are emerging as a useful tool for human researchers, with a crucial ability being proposing a research plan for a given research goal. In this work, we study how to train language models that generate better research plans by leveraging the vast corpus of existing research papers. To collect diverse training data, we automatically extract research goals and goal-specific grading rubrics from papers across domains. We then train models for research plan generation via reinforcement learning, with a frozen copy of the initial policy acting as the grader, using the rubrics to evaluate plans generated by the training policy. To validate this approach, we conduct a human study for machine learning research goals spanning 225 expert hours. The experts prefer plans generated by our finetuned Qwen3-30B-A3B model over the initial model for 70% goals, and over Grok-4-Thinking for 59.6% goals. To assess generality, we also extend our approach to goals from medical papers, and recent arXiv preprints, evaluating with a jury of frontier models. Our finetuning yields 12-22% relative improvements and significant cross-domain generalization, proving effective even in problem settings like medical research where execution feedback is infeasible. Overall, we demonstrate the potential of a scalable training recipe as a step towards improving general AI co-scientists.

Theory · Reinforcement Learning and Planning

Onno Eberhard, Claire Vernade, Michael Muehlebach

Theoretical properties of reinforcement learning algorithms are most commonly studied under the Markov assumption. This is unrealistic, as most environments encountered in practice are either partially observable, or require function approximation that restricts the agent to access non-Markovian state features. We consider the problem of learning an optimal reactive policy in a finite environment under deterministic observations (or equivalently, hard state aggregation). We introduce a new algorithm, _Committed Q-learning_, and prove almost sure convergence to the optimal reactive policy under an intuitive assumption we call _rewire-robustness_. This assumption is strictly weaker than the $q_\star$-realizability condition used in prior work. Our algorithm is a variant of classical Q-learning in which the behavior policy commits to a single action upon entering a feature, and only resamples actions when the observed feature changes.

Theory · Reinforcement Learning and Planning

Egor Cherepanov, Daniil Zelezetsky, Aleksandr Panov, Aleksey Kovalev

Pixel-based reinforcement learning agents often fail under purely visual distribution shift even when latent dynamics and rewards are unchanged, but existing benchmarks entangle multiple sources of shift and hinder systematic analysis. We introduce KAGE-Env, a JAX-native 2D platformer that factorizes the observation process into independently controllable visual axes while keeping the underlying control problem fixed. By construction, varying a visual axis affects performance only through the induced state-conditional action distribution of a pixel policy, providing a clean abstraction for visual generalization. Building on this environment, we define KAGE-Bench, a benchmark of six known-axis suites comprising 34 train-evaluation configuration pairs that isolate individual visual shifts. Using a standard PPO-CNN baseline, we observe strong axis-dependent failures, with background and photometric shifts often collapsing success, while agent-appearance shifts are comparatively benign. Several shifts preserve forward motion while breaking task completion, showing that return alone can obscure generalization failures. Finally, the fully vectorized JAX implementation enables up to 33M environment steps per second on a single GPU, enabling fast and reproducible sweeps over visual factors.

Deep Learning · Large Language Models

Mathieu Blondel, Michael Sander, Germain Vivier-Ardisson, Tianlin Liu, Vincent Roulet

Autoregressive models (ARMs) currently constitute the dominant paradigm for large language models (LLMs). Energy-based models (EBMs) represent another class of models, which have historically been less prevalent in LLM development, yet naturally characterize the optimal policy in post-training alignment. In this paper, we present a unified view of these two model classes. Taking the chain rule of probability as a starting point, we establish an explicit bijection between ARMs and EBMs in function space, which we show to correspond to a special case of the soft Bellman equation in maximum entropy reinforcement learning. Building upon this bijection, we derive the equivalence between supervised learning of ARMs and EBMs. Furthermore, we analyze the distillation of EBMs into ARMs by providing theoretical error bounds. Our results provide insights into the ability of ARMs to plan ahead, despite being based on the next-token prediction paradigm.

Deep Learning · Large Language Models

Maggie Ziyu Huan, Yuetai Li, Tuney Zheng, Xiaoyu Xu, Seungone Kim, Minxin Du, Radha Poovendran, Graham Neubig, Xiang Yue

Math reasoning has become the poster child of progress in large language models (LLMs), with new models rapidly surpassing human-level performance on benchmarks like MATH and AIME. But as math leaderboards improve week by week, it is worth asking: do these gains reflect broader problem-solving ability or just narrow overfitting? To answer this question, we evaluate over 20 open-weight reasoning-tuned models across a broad suite of tasks, including math, scientific QA, agent planning, coding, and standard instruction-following. We surprisingly find that most models that succeed in math fail to transfer their gains to other domains. To rigorously study this phenomenon, we conduct controlled experiments using math-only data with two widely-used methods: Reinforcement Learning (RL) and Supervised Finetuning (SFT) with detailed ablations. On top of the observation that RL-tuned models transfer better than SFT-tuned model, we identify on-policy fine-tuning as the key mechanism underlying cross-domain transfer, regardless of whether the training signal comes from RL or supervised learning. Latent-space representation and token-space distribution shift analyses reveal that off-policy SFT induces substantial representation and output drift, while on-policy RL preserves general-domain structure. Our results suggest a need to rethink the post-training recipes, particularly the reliance on off-policy SFT-distilled data to advance reasoning models.

Reinforcement Learning · Batch/Offline

Evgenii Opryshko, Junwei Quan, Claas Voelcker, Yilun Du, Igor Gilitschenski

Offline goal-conditioned reinforcement learning (GCRL) often struggles with long-horizon tasks, where errors in value estimation accumulate and produce unreliable policies. It is typically assumed that effective long-term planning is infeasible without specialized training. In contrast, our work demonstrates that existing GCRL policies can complete long-horizon tasks when combined with a lightweight, training-free planning wrapper. We find that standard goal-conditioned value functions encode locally consistent geometric structure sufficient for planning. Our approach, Test-Time Graph Search (TTGS), constructs a graph over the offline dataset and employs an adaptive subgoal selection strategy. To address unreliable value estimates during shortest-path search, we propose a novel mechanism that softly penalizes long-distance transitions. Our method incurs negligible computational overhead and requires no additional supervision or parameter updates. On the OGBench benchmark, TTGS unlocks latent capabilities in diverse base learners, boosting success rates on challenging locomotion tasks from near-zero to over 90\%, often matching or outperforming methods that require complex auxiliary training.