Group Relative Policy Optimization

Reinforcement learning with verifiable rewards is widely used for post-training reasoning models, but sparse outcome rewards make exploration difficult. A complementary approach is to optimize inference-time objectives such as pass@K and max@K directly, yet existing policy-gradient estimators for these objectives use different signals, baselines, and normalizations, making their relationships unclear. We study this issue through baseline design and advantage centering. Starting from the advantage estimator of a leading method in the field, we show that it is policy-gradient unbiased but yields a non-centered advantage. We then introduce a Leave-Two-Out baseline that preserves policy-gradient unbiasedness while making realized batch advantages exactly centered. The resulting method, MaxPO, has an efficient quadratic-time implementation and integrates naturally into group-based RL for LLM post-training. We further derive the canonical finite-batch advantage for max@K, providing a unified view of existing advantage estimators. Empirically, we verify that the L2O baseline reduces gradient variance and outperforms non-centered alternatives.

Reinforcement learning with verifiable rewards is ideal for multi-constraint instruction following, yet standard group-relative policy optimization (GRPO) becomes unstable under discrete, low-dispersion rewards, where within-group reward distributions are frequently homogeneous. We identify and formalize three pathologies of z-score group normalization in this regime: low-variance amplification, mean-centering blindness, and zero-variance collapse. To address them, we propose MDP-GRPO, which stabilizes learning through (1) multi-temperature sampling to increase reward dispersion, (2) dual-anchor advantages to restore gradients in homogeneous groups and stop mean-centering blindness, (3) prospect-theoretic shaping to bound updates and penalize violations based on Kahneman and Tversky's theory, and (4) asymmetric KL regularization. Evaluated on FollowBench, IFEval, and a curated multi-constraint dataset, MDP-GRPO outperforms standard GRPO, improving strict constraint satisfaction by up to 5.0% on Llama-3.2-3B. Our method also enables stable convergence with small group sizes while preserving general capabilities on MMLU and ARC.

Reinforcement learning with verifiable rewards (RLVR) often adopts GRPO-style group-relative updates, sampling multiple rollouts per prompt to construct normalized learning signals. However, merely increasing the number of rollouts does not reliably strengthen learning: under GRPO-style group normalization, per-rollout policy-gradient features can concentrate into a low-rank, signed geometry, causing substantial cancellation during aggregation and weakening the effective update. We address this failure mode with SALT, a Subspace-Adaptive geometry pLug-in componenT that uses sample-wise gradient geometry to reweight the coefficients of group-relative updates. SALT estimates a dominant shared subspace from the mini-batch Gram geometry, decomposes group-relative coefficients into shared and residual channels, and adaptively amplifies the residual channel when signed cancellation is severe. Across diverse reasoning-oriented RLVR benchmarks and model scales, SALT improves effective update geometry and performance without modifying the reward model or the rollout sampling procedure

We identify and formally characterize credit misassignment as a systematic failure mode of GRPO in tool-augmented multimodal search agents: its uniform broadcast of trajectory-level advantages to all tokens causes valuable tool-use steps in failing trajectories to be penalized no differently from valueless ones. We further empirically quantify the scale of this phenomenon. Over half of failing trajectories and failing tool-use actions exhibit correctable credit misassignment, demonstrating that the wasted training signal is both substantial and structurally exploitable. Building on this insight, we propose Tool-Aware Policy Optimization (TAPO), which exploits the parameter-determinism property of information-acquisition tools: similar call parameters define equivalent information-acquisition actions and should therefore share comparable action credit. TAPO constructs counterfactual witnesses within the current training batch and compensates misassigned negative credit via confidence-gated conservative advantage correction. It requires no additional annotation, models, or sampling, and introduces negligible computational overhead. Across multiple multimodal search benchmarks, TAPO delivers consistent, plug-and-play improvements over strong baselines for three mainstream RL algorithms (GRPO, GSPO, and SAPO). Our code and models will be publicly released upon acceptance.

Verifiable reward training has improved mathematical and coding reasoning, but these domains capture only part of step-by-step decision making. Many real-world tasks require finding a high-value feasible plan among many valid alternatives. We introduce OPT*, a scalable family of optimization-style tasks for training and evaluating LLM step-by-step optimization-like reasoning along a complexity axis: each task provides a feasibility checker and evaluator, while a complexity parameter expands the search space without requiring new human labels. This motivates studying these tasks in two regimes: (i) solver-guided online policy optimization, which uses a solver as a value oracle for partial states and applies rank-based reward shaping to reinforce better next steps, and (ii) search-based offline RL when such solvers are unavailable. Theoretically, we relate success in large search spaces to the information a reasoner extracts per unit of search budget. Empirically, we ablate the ingredients that make search efficient on OPT* and show that training on OPT* improves step-by-step optimization-like reasoning.

Group Relative Policy Optimisation (GRPO) has emerged as an effective reinforcement-learning algorithm for aligning language models on reasoning tasks, but it treats every token position and every sampled rollout symmetrically. We introduce two complementary extensions: (i) Adaptive-Horizon GRPO (AH-GRPO), which weights each token's policy gradient using a cumulative entropy-based discount that reduces the effective horizon when the model is uncertain, and (ii) Selective-Advantage AH-GRPO (SA-AH-GRPO), which applies this discounting only to negative-advantage rollouts, leaving positive-advantage, successful trajectories unattenuated. We evaluate standard GRPO with alpha = 0, AH-GRPO with alpha = 0.5, and SA-AH-GRPO with alpha = 0.5 on the GSM8K mathematical reasoning benchmark using both Qwen 2.5-1.5B-Instruct and Qwen 2.5-3B-Instruct fine-tuned with LoRA. On the 3B model, SA-AH-GRPO achieves Pass@1 = 0.858 at its peak at step 30 and maintains 0.846 at 180 steps, with training variance reduced to 0.0246, a 3.6 times reduction relative to GRPO while matching its peak accuracy. On the 1.5B model, SA-AH-GRPO achieves a peak Pass@1 of 0.686, improving over the zero-shot baseline of 0.637. Our analysis shows that asymmetric discounting preserves the full gradient signal on correct solutions, prevents entropy collapse, and substantially stabilises training, suggesting a principled inductive bias for reinforcement learning with verifiable rewards on structured generation tasks.

LLM agents operate in two distinct regimes: open-weight agents amenable to reinforcement learning (RL) and black-box agents whose behaviour must be controlled purely at test time. Although black-box agents are often backed by state-of-the-art proprietary LLMs, API-only access precludes parameter-level optimization, rendering most RL methods inapplicable. To address this limitation, we turn to a known equivalence between RL and Bayesian inference. We propose Agentic Monte Carlo (AMC) to directly sample from the optimal policy of a black-box agent rather than training it through RL. The optimal policy is a posterior over trajectories whose prior we define as the fixed black-box LLM agent. We employ Sequential Monte Carlo to sample from this posterior by learning a value function to steer the agent while leaving the underlying black-box model unchanged. We validate AMC on three diverse environments from the AgentGym benchmark, demonstrating significant improvements over prompting baselines and even outperforming Group Relative Policy Optimization (GRPO) as we scale the test-time compute of our method. AMC demonstrates the feasibility of performing principled RL-style optimization of black-box LLM agents. Code is available at https://github.com/layer6ai-labs/Agentic-Monte-Carlo

Mitigating social bias in Large Language Models (LLMs) presents a distinct alignment challenge: unlike verifiable tasks, bias lacks a single ground truth, creating a high-variance, subjective reward landscape. Previous preference-based fine-tuning methods have major trade-offs: Direct Preference Optimization (DPO) is limited by the lack of exploration inherent in offline training, while Proximal Policy Optimization (PPO) can lead to training instability due to potentially unreliable critic estimates. In this paper, we propose BiasGRPO, a framework using Group Relative Policy Optimization (GRPO) to stabilize alignment by normalizing rewards across a group of sampled completions. By substituting the value function with a group-relative baseline, our approach reduces instability while maintaining the exploration benefits of online training. We find that BiasGRPO outperforms DPO and PPO across multiple benchmarks, indicating its effectiveness. To adapt GRPO, we synthetically extend a dataset spanning multiple domains and contexts. We also create and release a custom bias reward model that effectively guides generation while being highly compute-efficient and avoiding knowledge degradation, providing a valuable resource that can be seamlessly integrated into multi-objective RLHF pipelines.

Reinforcement learning from verifiable rewards with GRPO is a standard approach for post-training reasoning LLMs. It remains sample inefficient. Each rollout is used for a single gradient update and then discarded. Naive replay is not well suited in this setting because LLM policies drift quickly per gradient step. Stored rollouts therefore become stale and can destabilize training. We propose a rollout-level replay buffer for GRPO that stores and samples individual rollouts rather than whole groups. The buffer bounds staleness through age eviction. Any rollout older than tau_max training steps is removed. The buffer also preserves on-policy data via fresh-anchored composition. Each batch keeps its fresh on-policy rollouts and then concatenates replay rollouts drawn separately from the buffer. We prioritize replay by per-rollout advantage magnitude and recycle individual rollouts whose advantages are large. Across three Qwen3-Base scales on five math benchmarks, our method outperforms GRPO and naive replay baselines. Gains are positive at every scale and reach +1.66 pp on the five-benchmark average at 4B. Under an AES metric that jointly measures accuracy and token efficiency, our method is the only condition with a positive margin over GRPO at every scale.

The framework of robust Markov decision processes (RMDPs) allows the design of reinforcement learning agents that satisfy performance guarantees under worst-case transition dynamics. Traditional RMDPs consider discrete-time dynamics and recently, sample-efficient policy gradient algorithms have been considered in this context. This paper investigates policy gradient algorithms within a continuous-time RMDP framework. Policy gradients and adversarial gradients are derived using pathwise and adjoint-based formulas for stochastic and ordinary differential equations. We propose double-loop optimisers to obtain linear convergence in the oracle-based setting and an $\tilde{\mathcal{O}}(\frac{1}{ε^2})$ sample complexity in the sample-based setting in an analysis which also derives novel tools for the framework of undiscounted total cost MDPs. Additionally, we propose mean-field optimisers as distributional optimisers with an $\tilde{\mathcal{O}}(\frac{1}{K})$ oracle-based convergence rate and an $\tilde{\mathcal{O}}(\frac{N^2}ε)$ sample complexity under $N$-particle approximation. The effectiveness of continuous-time policy gradient algorithms is confirmed for both optimisers on continuous-time RMDPs with neural ordinary differential equation dynamics.

Small open-weight models struggle at long-form creative writing: their generated stories either fall far short of the requested length, or their quality significantly degrades as length increases, especially when compared to frontier models. We present POLARIS (Policy Optimization with LLM-as-a-judge rewards and Anchored-Reference Injection for Storywriting), a lower-compute GRPO recipe with two key ingredients: a frontier LLM judge with a structured Story Quality rubric as the online reward, and human-reference injection (HRI), where a teacher-forced human-written story serves as a high-reward anchor within each GRPO group. By applying our training recipe to Qwen3.5-9B, using a dataset of approximately 1.4K prompt-story pairs derived from 100 short-story anthologies and 4 A100 GPUs, we obtain POLARIS-9B. Across five benchmarks spanning in-distribution and out-of-distribution prompts and rubrics, POLARIS-9B is competitive with much larger open-weight models while following length instructions more closely. A blinded human evaluation confirms that POLARIS-9B is preferred to the base Qwen3.5-9B and on par with Qwen3.5-27B. Despite training only on stories up to 4k words, POLARIS-9B preserves quality on prompts requesting stories up to 3 times the training length, a regime where most open-weight models degrade substantially in quality, length adherence, or both. More broadly, our results suggest that length generalization is a meaningful stress test for creative-writing models and a useful lens for distinguishing otherwise close models.

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

While Proximal Policy Optimization (PPO) demonstrates strong performance in stationary settings, we show that its standard optimization paradigm struggles in continual and non-stationary environments. The failure does not stem from insufficient model capacity or overly restrictive clipping. Instead, PPO performs persistent, directionally inefficient local updates, which indicates a lack of geometry-aware guidance for accumulating meaningful behavioral change and ultimately hindering transitions toward new behavior patterns. Although divergence-based regularization introduces partial geometric awareness, its monotonically increasing penalties implicitly discourage large policy deviations, even when such shifts are necessary for effective adaptation. To address this limitation, we propose Gaussian Trust Region Policy Optimization (GTR), which reshapes the trust region using a Gaussian kernel. The resulting constraint is bounded and non-monotonic, providing strong local stability while progressively relaxing under sustained high-advantage updates. To further improve robustness, we introduce a Mixture Gaussian Anchor that adapts to recent policy trajectories, reducing variance induced by stale references. GTR is architecture-agnostic and achieves strong performance across games, simulated robotic control, open-world exploration, and language model post-training. These results demonstrate that geometry-aware trust-region design can be a promising direction for robust reinforcement learning in complex non-stationary environments. Our code is available at https://anonymous.4open.science/r/GTR_demo/README.md.

Large scale GPU-parallel reinforcement learning has changed what can be trained in robot simulation, yet most systems still optimize one specialist policy per task. We propose a construction methodology for turning structured manipulation task families into GPU-parallel multi-task RL benchmarks, and instantiate it as MT-Libero using LIBERO assets and task predicates in Isaac Lab. The resulting benchmark supports simultaneous reinforcement learning over heterogeneous task suites with parallel rendering, physics randomization, and state-input or visual-input policies. To make such training practical under sparse success signals and limited prior data, we further propose DGPO, an on-policy demonstration guided method that combines importance weighted PPO with adaptive behavior cloning on matched demonstration actions. DGPO enables a tunable preference toward demonstrated task distributions, outperforming both prior-free RL and existing demonstration-based methods while preserving the stability and online improvement benefits of on-policy PPO.

Group Relative Policy Optimization (GRPO) and its variants, originally developed for Large Language Models (LLMs), have recently been applied to Multimodal LLMs and produced strong results. However, their coarse-grained holistic credit assignment from a single scalar advantage underfits vision-language (VL) tasks, where outputs are often long-form responses grounded in semantically rich images. To address this limitation, we exploit a structured signal that single-scalar formulations discard: the natural segmentation of long-form VL outputs. Concretely, we propose Segment-Decomposed GRPO (SD-GRPO), which z-normalizes verifiable per-segment rewards across the rollout group, yielding a vector of per-segment advantages in place of a single scalar. We evaluate SD-GRPO across three settings spanning controlled and real-world long-form VL generation, organized by increasing semantic entanglement across segments. On a controlled multi-panel dense-captioning task constructed from DOCCI, where segments are semantically independent, SD-GRPO consistently outperforms the GRPO baseline, with larger gains at higher segment counts. Extending to a controlled multi-chart long-form VQA task constructed from MultiChartQA, we show both theoretically and empirically that rollout-level rewards suffer from cross-segment credit misattribution that scales with output length. On a real-world scientific figure captioning task on the MMSci dataset, where subfigure captions share context across the figure, blending holistic and per-segment rewards further improves on both, suggesting per-segment normalization alone is insufficient when segments are semantically entangled. Finally, by integrating SD-GRPO into Dr. GRPO, we confirm that it can be applied to any GRPO framework with minimal implementation overhead to enhance long-form VL generation.

Recent advances in language models have established reinforcement learning as the primary paradigm for eliciting self-correction and long-chain reasoning. While group relative policy optimization (GRPO) offers superior scalability by eliminating the critic network, deploying it on a central infrastructure entails collecting a large volume of data from distributed owners, which poses significant privacy risks. To address these concerns, we introduce federated GRPO (FGRPO), a framework designed to decentralize the fine-tuning of reasoning models across heterogeneous data owners. To effectively mitigate the instability caused by divergent reward scales across heterogeneous tasks, FGRPO incorporates an adaptive aggregation mechanism based on relative performance gain. By characterizing each client's improvement relative to its personalized historical baseline, the framework dynamically prioritizes effective learning trajectories regardless of local task difficulty. FGRPO ensures robust convergence on non-IID data while preserving data privacy.

On-policy self-distillation, where a language model conditions on privileged context to supervise its own generations, is a promising source of dense supervision for sparse-reward reinforcement learning. Actually, it can be instantiated as an auxiliary full-vocabulary student-to-teacher reverse Kullback-Leibler divergence loss. We therefore propose SDPG, a self-distilled policy-gradient framework that combines group-relative verifier advantages with normalized standard deviation, exact full-vocabulary on-policy self-distillation, as well as reference-policy KL regularization. Empirically, SDPG improves stability and performance over RLVR and self-distillation baselines. The code is available at https://github.com/lauyikfung/SDPG.

Recent developments in Large Language Models (LLMs) have showcased impressive reasoning capabilities, with Reinforcement Learning with Verifiable Rewards (RLVR) being a promising enhancement strategy. However, existing reward mechanisms are constrained to the outcome-level correctness and lack explicit signals to guide the model to consider diverse solutions. In contrast, human problem solving typically involves evaluating multiple potential approaches and selecting the most reliable solution, a cognitive process that current RLVR frameworks do not explicitly incentivize. Inspired by this, we propose Hint-Guided Diversified Policy Optimization (HDPO), allowing the model to first list all potential candidate solution outlines as hints and then select the most reliable one for further reasoning. HDPO comprises two stages of Cold Start for Structured Reasoning and Hint-Guided Diversified Reinforcement Learning to incentivize the model to generate diverse and reliable solutions following the ``propose-select-think'' trajectory. Experimental results show that HDPO effectively boosts LLM reasoning and enhances the diversity of candidate solutions as well as the LLM's ability to identify reliable solutions.

Vision-Language-Action (VLA) models offer a promising end-to-end paradigm for unmanned aerial vehicles (UAVs) to accomplish complex tasks specified by fine-grained instructions. However, standard supervised fine-tuning (SFT) suffers from data scarcity, limited generalization, and weak supervision for nuanced and complicated human intents. Reinforcement fine-tuning offers a natural way to mitigate these challenges and align policy behaviors with human intents through designable feedback, but applying it to aerial navigation remains challenging due to inefficient exploration in expansive continuous spaces. To address these challenges, we introduce an efficient reinforcement learning (RL) framework for VLA-based aerial navigation. At its core, we propose EG-GRPO (Expert-Guided Group Relative Policy Optimization) to augment online rollouts with few-shot expert data. Additionally, we design a heterogeneous pipeline enabling parallel simulation and inference, which reduces rollout time by 43.5%. Across multiple tasks specified by complex human intents, EG-GRPO improves the success rate to 2.13x that of the SFT baseline, while improving intent alignment performance by 60.9%. These results demonstrate that our framework can move aerial navigation toward precise intent-aligned flight.

Synchronous reinforcement learning methods such as Group Relative Policy Optimization (GRPO) provide stable and reproducible on-policy training, but they are highly vulnerable to stragglers, a single unusually long rollout can delay reward computation and parameter updates for the entire group. This problem becomes more severe as group size increases, creating a tension between the benefits of larger groups and the wall-clock cost of synchronization stalls. We propose Straggler-Aware Group Control (SAGC), a dynamic group-size controller that adapts the training group online based on observed rollout behavior. SAGC formulates group-size selection as an online constrained optimization problem, seeking to retain the benefits of larger groups while controlling the long-term rate of straggler events. Across synchronous GRPO and DAPO training, and on top of both vanilla and strong engineered baselines, SAGC consistently reduces straggler incidence and improves wall-clock efficiency while achieving competitive or better training reward. We further show that these gains transfer to final model quality: SAGC is competitive with or better than the strongest static group-size baseline on downstream reasoning benchmarks, and often produces shorter outputs without any explicit length penalty. These results position dynamic group control as a practical way to make synchronous on-policy RL more efficient and robust.

Post-training via Group Relative Policy Optimization (GRPO) has emerged as a powerful paradigm for aligning flow-based generative models with human preferences. However, the iterative denoising nature of flow models incurs substantial costs when generating group rollouts for policy-gradient updates, compelling existing methods to train with extremely few denoising steps. This temporal sparsity severely restricts preference optimization: reward feedback can only reach a handful of stages per trajectory, leaving the vast majority of intermediate denoising steps without direct supervision and thus compromising alignment granularity. To address this, we propose Pave-GRPO, which reformulates the GRPO objective through Principled average velocity decomposition. Rather than generating expensive high-step rollouts, we maintain efficient few-step group sampling but decompose each coarse transition into an equivalent ensemble of finer sub-trajectories spanning multiple intermediate timesteps. This propagates reward feedback to a denser set of temporal stages for more comprehensive preference alignment without additional generation cost. This design offers two benefits: (i) zero-cost horizon expansion: through the direct reuse of piece-wise group samples and their associated rewards, Pave-GRPO significantly broadens the effective optimization scope under fixed sampling budgets; and (ii) comprehensive temporal supervision: by equivalently decomposing an instantaneous velocity target into a multi-timestep ensemble, it distributes reward signals across more intermediate stages of the denoising process, enabling finer-grained and more thorough preference optimization. Extensive experiments validate that Pave-GRPO effectively advances preference alignment across different reward settings, offering comprehensive performance enhancement.

Reinforcement learning with verifiable rewards (RLVR) has emerged as a powerful paradigm for enhancing the reasoning capabilities of large language models (LLMs). However, its effectiveness is substantially hindered by the prevalence of ineffective training data: many sampled prompts yield response groups that are either entirely correct or entirely incorrect, resulting in zero-variance rewards and limited learning signals. Recent state-of-the-art methods address this issue through extensive LLM rollouts to filter ineffective samples, but at the cost of considerable computational overhead. Alternative approaches, including predictive sampling and trajectory replay, aim to improve data efficiency but often remain insufficient and may introduce additional issues such as systematic bias or suboptimal constraints. To address these limitations, we propose Group Prioritized Off-Policy Optimization (POPO), a simple yet effective framework that fully exploits effective training batches without additional rollout overhead. POPO comprises two key components: prioritized group replay and decoupled off-policy optimization. The former replaces ineffective on-policy groups with effective off-policy groups via a recency-based replay mechanism that jointly considers sample quality and the degree of off-policiness. To further mitigate the off-policy gap, POPO employs decoupled importance sampling to correct off-policy bias while maintaining stable policy updates under consistent trust-region constraints. Empirical evaluations across diverse reasoning tasks, including mathematics, planning, and visual geometry, demonstrate that POPO substantially accelerates RL finetuning and achieves strong reasoning performance with significantly fewer rollouts.

Inductive generalization is a framework for reinforcement learning (RL) generalization in which inductively related task instances admit inductively related policies. Prior work captures this structure via a higher-order policy-evolution function learned directly with RL, but suffers from poor training scalability: as training tasks grow, aggregated reward feedback becomes noisy and conflicting, destabilizing training and weakening generalization. We propose DIBS, a decoupled behavioral cloning approach that separates learning task-specific policies from learning the evolution function. We first learn individual teacher policies per task via standard RL, then fit the evolution function via behavioral cloning on teacher-labeled state-action pairs. This replaces noisy reward aggregation with dense, stable supervision. DIBS achieves significant improvements in both training stability and zero-shot generalization against existing RL and meta-RL algorithms.

This paper investigates the optimization mechanisms of non-convex Model Predictive Control (MPC) using the Maximum Entropy Differential Dynamic Programming (ME-DDP) framework. Navigating non-convex cost landscapes induced by nonlinear dynamics, multiple obstacles, etc. remains a fundamental challenge in robotics, where gradient-based methods frequently converge to suboptimal local minima. We demonstrate a dual-step optimization mechanism designed to overcome these traps. (1) an initial phase of using DDP to exploit the gradient of the cost landscape, followed by (2) disruption of the optimization via sampling from policies characterized by the inverse Hessian of the action-value function. We provide a rigorous analysis of this sampling mechanism of three ME-DDP variants: Unimodal Gaussian ME-DDP, Multimodal Gaussian ME-DDP, and Stein Variational DDP. Furthermore, with navigation tasks of four robotic systems under cluttered environments, we conduct extensive benchmarking of three variants of the ME-DDP, against deterministic DDP, and one of the most successful sampling-based schemes, Model Predictive Path Integral (MPPI) control with three policy parameterizations and update laws that correspond to those of ME-DDPs. The results show that in low-dimensional systems where the cost landscapes are relatively simple and local information is sufficiently representative, our framework consistently outperforms MPPIs. In high-dimensional systems, MPPI can occasionally discover aggressive maneuvers that enable it to steer the systems faster than DDP-based methods, whereas our method maintains a higher, more stable success rate. Finally, we validate the practical efficacy of the framework through hardware experiments with a quadrotor navigating a dense, non-convex obstacle field, confirming the robustness of the proposed framework for real-world deployment.

Recent diffusion transformers have demonstrated strong image synthesis capabilities but remain inefficient to train due to weak alignment between generative and discriminative representations. While representation alignment frameworks such as REPA improve convergence by aligning noisy denoising features with pretrained visual encoders, their externally supervised alignment loss is static and lacks adaptivity during training and inference. Existing methods rely on fixed cosine alignment or contrastive objectives, which cannot dynamically balance representation consistency and generation quality, resulting in limited discriminative benefit and failing to optimize alignment in a task-adaptive manner. To address this, we propose VRPO, a reinforcement-based optimization strategy that replaces REPA's static alignment loss with a generative representation policy optimization objective. Instead of enforcing a fixed similarity constraint, VRPO treats representation alignment as a reward-guided process: the model receives adaptive rewards based on generation fidelity, perceptual quality, and semantic coherence between the diffusion features and pretrained visual embeddings. This formulation enables the generator to continuously refine its internal representations toward semantically meaningful directions while improving image quality. Our VRPO-driven training seamlessly integrates into diffusion transformers, introducing negligible computation cost and preserving full compatibility with SiT and DiT architectures. Extensive experiments on ImageNet-256x256 demonstrate that our VRPO-Alignment substantially enhances both convergence and fidelity, achieving up to +1.8 FID improvement and 2.3x faster training compared to REPA under identical compute budgets.

Deep reinforcement learning (RL) offers a promising route to real-time power grid operation, yet large neural policies are costly to evaluate, hard to deploy on constrained hardware, and opaque to operators. We ask whether a Proximal Policy Optimization (PPO) agent for grid topology control can be compressed into compact tree-based surrogates without losing operational performance. A PPO teacher is trained on Grid2Op's standard 14-bus environment with a stability-oriented reward, using stress-focused data collection on critical, high-loading states. The policy is then distilled into a decision tree and a random forest. Across held-out validation episodes, both surrogates exceed the teacher in mean reward and survival length at a fraction of the inference cost. The decision tree shows high exact-action agreement with the PPO argmax and near-complete agreement within its top-ranked actions, while remaining small enough to be inspected directly. Feature-importance analysis reveals a representational shift: the PPO policy relies mainly on line-loading signals, while the distilled tree is driven primarily by bus-topology variables. These results suggest that stress-focused distillation can convert a black-box neural controller into a lightweight, auditable rule-like surrogate suited for real-time deployment, while also surfacing risks tied to deterministic actions and topology-specific generalization.

In semiconductor manufacturing, lithography projects circuit layouts onto silicon wafers through an optical mask. As circuit features shrink below the wavelength of light, optical diffraction causes the printed patterns to deviate from their intended layouts. Inverse Lithography Technology (ILT) addresses this challenge by generating optimized masks that enhance the fidelity of pattern transfer onto wafers. While ILT resembles an image synthesis task, its reliance on explicit physical metrics for mask evaluation limits the applicability of existing generative models. We introduce LithoGRPO, an ILT framework that integrates the flow-matching paradigm with GRPO-based reinforcement learning (RL) fine-tuning, enabling efficient exploration of diverse masks for a given target layout. Unlike purely generative or optimization-based approaches, RL in LithoGRPO exploits the explicitly defined, physics-based reward function of ILT, enabling optimization under complex, process-aware constraints. To the best of our knowledge, this is the first framework that unifies flow matching and RL for mask optimization. To improve RL sampling efficiency, we propose a fast shot-counting algorithm for manufacturability evaluation, achieving over 130x speedup while preserving the mask ranking of the traditional shot-count metric. Extensive experiments demonstrate that LithoGRPO achieves state-of-the-art performance over both optimization-based and learning-based methods, while maintaining efficient mask generation.

Reinforcement learning with verifiable rewards (RLVR), especially Group Relative Policy Optimization (GRPO), has been widely used to improve reasoning in large language models. However, outcome-level rewards provide only sparse supervision, and group-relative advantages vanish when all sampled trajectories for a prompt are either correct or incorrect. On-Policy Self-Distillation (OPSD) offers dense token-level guidance, but its token preferences are not necessarily aligned with trajectory correctness; empirical diagnostics show that OPSD signals behave differently on correct and incorrect rollouts, with teacher-positive and teacher-negative gap signals exhibiting different noise profiles. These diagnostics are conducted under an OPSD-style privileged teacher context for analysis only, whereas CAST training uses answer-free self-teacher scoring.Motivated by these observations, this work proposes CAST, an answer-free self-distillation method for GRPO-style RLVR. CAST keeps the verifier-grounded GRPO objective, but uses a stop-gradient self-teacher to shape token-level advantages according to trajectory correctness. Unlike prior self-distilled RLVR methods, CAST does not require reference-solution-conditioned teacher scoring, keeps the self-teacher log-probability gap active throughout training, and applies bidirectional local advantage sign reversal: teacher-negative tokens in correct trajectories can receive negative token-level advantages, while teacher-positive tokens in incorrect trajectories can receive bounded positive local advantages. For zero-variance all-correct and all-wrong groups, CAST assigns bounded sign-constrained base advantages, so these otherwise zero-gradient groups can contribute verifier-signed token feedback. Experiments on mathematical reasoning show that CAST improves RLVR training while retaining a lightweight, verifier-grounded trajectory-level objective.

Reinforcement Learning with Verifiable Rewards is an effective route for post-training to strengthen the reasoning capability of large language models. However, as training proceeds, the learning signal can collapse thus makes the training gain become marginal and ineffective. Specifically, a growing fraction of prompts' rollouts become advantage-degenerated: all the self-generated rollouts show verified-success, making the standard deviation over their rewards be zero; accordingly each rollout's advantage becomes degenerated (zero) as well. Given such rollouts' advantages, the policy-gradient for model optimization eventually vanishes, capping the training performance. We argue that some of these rollouts still contain valuable learning signals but unfortunately omitted with the existing RLVR methods. In this paper, inspired through analyzing the entropy pattern behind golden trajectories produced by external expert models, we propose EchoRL for better exploiting the advantage-degenerated rollouts to further improve the training performance. EchoRL is a lightweight module that first identifies an EchoClip from verified-success rollouts based on their step-level entropy values, and then feeds this clip back as an auxiliary supervision signal in the RL objective. Extensive experiments across 10 benchmarks, 5 LLM backbones, and 4 popular RLVR post-training methods demonstrate that EchoRL consistently improves RLVR post-training with minimal overhead.

Bidding in repeated auctions is a central challenge for reinforcement learning (RL), combining continuous control with the strategic complexities of digital advertising. While policy gradient and value-based methods seem well-suited for these settings, they often struggle with the discontinuous, "cliff-like" nature of auction reward landscapes. In a first-price auction, for example, a bidder receives zero reward until they cross a specific threshold, after which the reward decreases as the bid increases. This creates a landscape of flat, zero-reward regions separated by sharp boundaries. We identify a fundamental failure mode in this setting termed "zero collapse." We show that stochastic exploration and gradient-based updates can cause policies to overshoot optimal high-reward regions and enter flat, zero-reward regimes. Once there, the lack of an informative gradient signal makes recovery extremely sample-inefficient, effectively trapping the agent. We find that actor-critic methods are particularly susceptible, as biased value estimates can accelerate this movement toward unstable regions. Our contributions include: (1) a mechanistic explanation of how discontinuous rewards lead to vanishing signals and zero collapse; (2) an analysis of the interaction between policy stochasticity and step size; and (3) an empirical demonstration of this phenomenon across REINFORCE and actor-critic variants. We propose practical mitigation strategies involving initialization and architectural choices to improve stability. Finally, we introduce a formal RL framework for auction environments highlighting their unique structural properties.

Post-training for reasoning models typically combines supervised fine-tuning with reinforcement learning from verifiable rewards, most commonly with GRPO. However, this algorithm suffers from sparse rewards, limited exploration, and mode collapse. Building upon recent works on self-distillation, we propose Feedback Distillation, a training method where the model is trained to match, at the token level, its own distribution conditioned on privileged feedback produced by a language model. Feedback Distillation offers token-level supervision and can inject external knowledge. Evaluating our method for Lean4 theorem-proving, we find that Feedback Distillation maintains greater diversity in generated trajectories than GRPO, yielding higher policy entropy and better pass@k scaling. The two methods are complementary: initializing GRPO from a Feedback Distillation checkpoint outperforms either method alone. All in all, our results suggest a promising avenue to improve post-training for complex reasoning.

Language models fine-tuned with reinforcement learning typically optimize for task reward, ignoring multi-agent strategic structure. Because these agents condition on natural language game-state descriptions and emit actions through free-form generation, strategic failure modes -- exploiting weaker opponents, coordinating on harmful equilibria, and externalizing costs are inseparable from the language interface itself. We propose Safe Equilibrium Policy Optimization (\sepo{}), a training objective that augments expected payoff with explicit penalties for exploitability, collusion risk, and externality cost. We implement \sepo{} as a reward signal for Group Relative Policy Optimization (GRPO), applied to Gemma4 E4B-it and Qwen3.5-4B after supervised fine-tuning (SFT). Evaluated across five strategic domains: Iterated Prisoner's Dilemma, repeated auctions, two negotiation variants, and Kuhn Poker. \sepo{} achieves zero exploit-pool advantage in Kuhn Poker for both models, outperforms the base model on safety in four domains, and corrects the over-cooperative behavior introduced by SFT. In negotiation, \sepo{} achieves a positive-safety outcome and only the positive normalized relative advantage of any negotiation configuration. Ablation experiments confirm that per-rollout exploit computation is necessary: a shared constant penalty cancels in GRPO advantage normalization (constant control-variate property), producing zero gradient. To support further research in strategic safety for agents, we release our \href{https://anonymous.4open.science/r/sepo-2668/README.md}{code} and SFT datasets.

We identify a new dimension for enhancing rollout diversity in Group Relative Policy Optimization (GRPO) for LLMs. While GRPO relies on diverse rollouts, prevailing strategies primarily increase diversity by injecting more token-level randomness, which may introduce step-wise noise and lead to incoherent trajectories. We uncover that smaller models within the same model family inherently exhibit higher policy-level diversity, indicated by their superior pass@k relative to larger counterparts as sample counts increase. Unlike token-level noise, this diversity is temporally correlated, preserves logical consistency, and provides structured exploration signals for gradient estimation. We thus propose S2L-PO (Small-to-Large Policy Optimization), a framework that leverages fixed small models as natural explorers to train larger models. To balance exploration and exploitation, we design a progressive annealing strategy that transitions from offline small-model rollouts to the large learner's own sampling. This shift elegantly avoids mid-training performance drops caused by the small model's capacity limits, achieving faster convergence and unlocking a higher performance ceiling. S2L-PO improves accuracy on diverse mathematical reasoning benchmarks (e.g., +8.8% on AIME 24 using a 1.7B explorer to guide the 8B model) while reducing rollout compute.

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

Recent work has analyzed the convergence of first-order methods under non-uniform smoothness assumptions that better model the loss landscape in machine learning tasks. We generalize this assumption to objectives whose curvature is an affine function of the objective value. This property is satisfied by a broad class of problems, including logistic regression, generalized linear models with a logistic link function, softmax policy gradient in reinforcement learning, and a class of neural networks. Under this assumption and gradient domination conditions, we establish a general convergence rate for the steepest descent method, and deterministic, diagonal variants of RMSProp and Adam. Our results imply that for logistic regression on separable data and the softmax policy gradient objective, sign GD converges linearly and is provably faster than GD. Furthermore, we show that for a class of two-layer neural networks on separable data, RMSProp and Adam can converge at a linear rate with a constant step-size and momentum parameter. Finally, we present a lower bound demonstrating that, under our assumption, RMSProp and Adam are provably faster than AdaGrad, AMSGrad, gradient descent, and heavy-ball momentum.

Group Relative Policy Optimization (GRPO) is an effective recipe for training reasoning models with verifier-based outcome rewards, but its supervision is sparse: when all sampled trajectories for a prompt receive the same verifier reward, the group-relative advantage collapses to zero and learning stalls. Outcome-only rewards also provide no step-level credit assignment, limiting exploration and making it harder to learn robust reasoning. We present VeriGate (Verifier-Gated Step-Level GRPO), a verifier-gated extension of GRPO that addresses these limitations with three design choices. First, VeriGate keeps the verifier in charge whenever verifier rewards induce a meaningful preference among sampled trajectories, and uses process supervision only when verifier rewards are degenerate. Second, instead of collapsing Process Reward Model (PRM) step scores into a single trajectory reward, VeriGate converts them into future-cumulated rewards to assign continuation-aware credit. Third, VeriGate transforms these rewards into group-normalized token-level advantages, restoring informative gradients and fine-grained credit assignment while remaining less susceptible to reward hacking than methods that optimize aggregated PRM scores. Empirically, training on MATH with 1.5B and 7B Qwen2.5-Instruct models and evaluating on six reasoning benchmarks, VeriGate improves average accuracy by about 20% and 12% for 1.5B and 7B models respectively, substantially reduces zero-gradient failures, decreases reward-hacking behavior, and improves reasoning quality relative to outcome-only GRPO and PRM-as-outcome baselines.

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

When a large language model under reinforcement learning commits a wrong reasoning step early in a trajectory, standard algorithms force it to keep generating until the maximum horizon, spending compute on tokens that never receive positive reward and polluting advantage estimates with post-failure noise. We propose ESPO (Early-Stopping Proximal Policy Optimization), which detects trajectory failure on-the-fly and terminates rollouts early. At each generation step, ESPO computes a surrogate regret using only the logits already computed during sampling, and terminates when the smoothed cumulative regret significantly exceeds its estimated values. Truncated trajectories are treated as absorbing failure states with a terminal reward, concentrating negative temporal-difference (TD) errors near the detected failure step without any additional reward model or human annotation. On DeepSeek-R1-Distill-Qwen-7B trained for mathematical reasoning, ESPO surpasses PPO on AIME2024 (46.28% vs. 45.25%), AMC2023 (85.83% vs. 82.94%), and MATH-500 (87.42% vs. 85.43%), while saving more than 20% rollout tokens cumulatively.

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

Recent reinforcement learning (RL) post-training approaches primarily optimize the final output policy using sparse outcome-level rewards, while largely overlooking predictive signals encoded in intermediate representations. In this paper, we introduce a new paradigm called on-policy internal self-distillation and propose the OISD framework, which improves reasoning by transferring on-policy predictive signals from the final layer to intermediate representations. During rollout and Group Relative Policy Optimization (GRPO) optimization, the final layer acts as both the policy and a detached internal teacher for selected intermediate layers, which are guided to align with it through two complementary mechanisms: logit alignment, which transfers high-level reasoning behaviors (how to think), and attention alignment, which enforces consistent attention patterns (where to look) from the final layer to the selected intermediate layer, both without requiring external privileged information. Our OISD, together with GRPO, employs signed advantage-weighted Jensen--Shannon alignment to distill informative intermediate representations while preserving policy consistency under a unified acting policy. Experimental results demonstrate the effectiveness of OISD, with substantial and consistent improvements over strong reasoning RL baselines across four mathematical reasoning tasks. The code will be released at https://github.com/THE-MALT-LAB/OISD

Large language models increasingly rely on either reinforcement learning or multi-agent prompting to improve reasoning, yet these two paradigms remain difficult to combine. Directly applying single-agent reinforcement learning to multi-turn multi-agent systems faces following dilemmas: i) Sparse rewards, role-level free-riding and excessive training overhead. ii) Agents only imitate to collaborate. iii) Fixed collaboration protocol falls into oscillating local optimum. We introduce TRACER, a turn-level reinforcement framework for cooperative multi-LLM reasoning. TRACER separates collaborative decision making into a controller-regret layer, where controllers learn whether the agents should speak or skip the current round through regret matching, and a generation-credit layer, which optimizes proposer and reviewer utterances with role-specific GSPO rewards. This design i) assigns credit at the level of both action modes and generated utterances, thus avoiding free-riding and sparse rewards. We only expand the choices made by the controllers, thus greatly reducing computational cost of training. Moreover, ii) agents acquire collaborative capability as they learn when to utter and what to speak. Finally, iii) by designing binary actions ingeniously, we extend classical game theory established for finite action spaces to deep learning, thus achieving mathematically rigorous convergence. We train all local RL-style methods on the GSM8K training split and evaluate on held-out GSM8K, MATH500, and GPQA-Diamond to measure in-domain accuracy, cross-benchmark generalization, inference cost, and correction-preservation behavior. The resulting framework provides a compact and reproducible testbed for studying learned collaboration policies beyond fixed debate, voting, or aggregation protocols. Code is available at https://github.com/Shark-Forest/TRACER.

Proactive Recommender Systems (PRSs) aim to guide user preference shift toward target items by generating paths of intermediate recommendations. Reinforcement learning (RL) provides a principled framework for optimizing such sequential decision tasks, as path rewards can naturally capture both short-term acceptance and long-term guidance effectiveness. However, naively applying policy gradients to PRS results in deficient gradient estimation. We identify two deficiencies: (1) path-level rewards decompose into step-level rewards with positive mean, creating a length-dependent bias that causes gradients to favor path extension over meaningful exploration; (2) weighting each step by the entire path-level reward ignores the decomposition structure, leading to high gradient variance. To rectify these two deficiencies, we propose an effective RL framework ProRL with two novel mechanisms for proactive recommendation. First, Stepwise Reward Centering subtracts expected rewards to neutralize length-dependent bias, ensuring that path extension yields zero expected gradient signal. Second, Position-Specific Advantage Estimation leverages the reward decomposition structure to compute step-dependent baselines, reducing gradient variance. Together, these mechanisms yield policy gradients that precisely target path quality. Our experiments on three real-world datasets demonstrate that ProRL significantly outperforms state-of-the-art PRSs. Our code is available at https://github.com/hongruhou89/ProRL.

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

Group Relative Policy Optimization (GRPO) is widely used for training reasoning models, but updating all sampled completions in each group incurs substantial cost and can reinforce verbose reasoning trajectories. In this paper, we study whether all completions provide equally useful update signals in GRPO-style reasoning RL. Our gradient-similarity analysis shows that, within the same prompt group, same-class completions often induce highly similar update directions, whereas correct-incorrect pairs provide more distinct contrastive signals. Motivated by this observation, we propose Binary Prefix Policy Optimization (BPPO), which uses the shortest correct completion and the shortest incorrect completion as a compact update unit while preserving full-group advantage normalization. BPPO further improves efficiency with adaptive completion scheduling and prefix-focused optimization; by updating only response prefixes, it avoids reinforcing redundant suffixes and encourages more concise responses. Experiments on GSM8K, MATH, and Geo3K show that BPPO achieves up to 6.08x speedup over GRPO while maintaining competitive accuracy, and reduces mean response length by approximately 30-50% without modifying the reward with an explicit length penalty.

Claim verification splits between end-to-end classifiers that are accurate but yields no inspectable traces, and decomposition-based methods produce inspectable traces but lag performance on benchmark datasets. We propose DecomposeRL an accurate claim-verifier that produce inspectable traces. DecomposeRL frames decomposition as an RL policy trained with GRPO and a multi-faceted reward ensemble, enabling both fully supervised and semi-supervised learning from unlabeled claims. DecomposeRL addresses the prohibitive training cost of GRPO with a data-curation funnel that distills 115K fact-verification claims into a compact, learning-signal-dense subset of 5K claims. We show that a DecomposeRL-7B policy trained with full supervision on only ~5K curated claims achieves 86.3 in-domain and 69.8 out-of-domain balanced accuracy across 11 claim-verification benchmarks containing biomedical, political, scientific, and general-domain claims. Despite being 4x smaller, it matches 32B baselines and GPT-4.1-mini, and it further outperforms baselines in a semi-supervised setting with only 10% labeled claims data. Code, data, and models are available at https://dipta007.github.io/DecomposeRL

Audio and omni-modal large language models exhibit impressive cross-modal reasoning capabilities. However, applying standard reinforcement learning post-training algorithms to these models exposes a critical structural vulnerability: methods like GRPO apply uniform policy gradients across all tokens, ignoring their unequal dependence on the non-text source modality. This exacerbates late-stage modality collapse during extended chain-of-thought generation, where models progressively abandon the primary source signal in favor of compressed textual priors, leading to confident but ungrounded hallucinations. To address this, we introduce Modality-Aware Policy Optimization (MAPO), a novel dual-branch reinforcement learning framework. First, MAPO dynamically concentrates the policy gradient on modality-critical tokens using a modality relevance mask, which is derived from the cross-modal differential entropy between an audio-ablated reference and the multimodal policy. Second, it integrates an auxiliary attention loss branch that applies a targeted, temporally scaled penalty to the model's internal attention distributions. This ensures the model actively sustains cross-modal grounding deep into the reasoning trace. Evaluations on complex audio reasoning benchmarks demonstrate that MAPO substantially improves long-horizon reasoning fidelity and multimodal instruction following, achieving highly competitive performance and setting new state-of-the-art results on several key benchmarks among open-weight models. By relying strictly on native statistical signals rather than domain-specific inductive biases, MAPO offers a promising foundation for mitigating epistemic collapse across diverse multimodal systems.

A modern post-training pipeline often writes one symbol for its policy, pi_theta, while evaluating it through two different programs: a training kernel optimized for autograd and an inference kernel optimized for low-precision, fused, dynamically batched serving. In finite precision, these kernels can induce different distributions at identical weights, with the gap concentrated on slices that aggregate benchmarks under-represent. This paper proposes kernel contracts: a contract-first framework for specifying acceptable divergence between K_train and K_inf. A contract C = (N, S, R, O, Pi) combines numerical, statistical, runtime, and observability clauses with an escalation policy from violations to routing actions. We derive a chain of bounds from logit drift to total-variation distance to bounded reward drift, and specialize it to RL post-training, where per-token importance-ratio drift yields a bound on policy-gradient bias under explicit support and norm assumptions. We also describe a four-stage promotion pipeline, online routing loop, and minimal YAML DSL for contract artifacts. This is a framework and vocabulary paper; we do not report production-scale empirical validation.

Few-shot prompting provides an effective way to adapt auditory large language models to low-resource tasks such as children's speech recognition. However, most auditory large language models are not explicitly trained to perform inference in this demonstration-conditioned format, limiting the extent to which they can benefit from few-shot prompting. To address this limitation, we introduce Few-Shot Aware GRPO (FSA-GRPO), an RL-based post-training recipe that uses a specially designed reward to encourage the model to leverage few-shot demonstrations, thereby strengthening its few-shot adaptation ability. Notably, training with only high-resource adult ASR data improves the model's general few-shot adaptation ability, yielding gains not only in children's speech recognition but also in speech translation and audio understanding. We further study data selection and auxiliary reward weighting to identify an effective training recipe. Our experiments show that when in-domain data are unavailable or cannot be used for training, FSA-GRPO is more effective than direct tuning on related out-of-domain data.

While rule-based reinforcement learning has recently catalyzed explicit reasoning in multimodal models, tactile reasoning remains largely underexplored. Existing tactile-language models primarily rely on supervised or contrastive objectives, which limits their capacity to ground predictions in physical evidence or rectify misleading visual priors. Tactile reasoning introduces two modality-specific challenges: the ordinal nature of physical attributes (e.g., hardness, roughness) and the cross-sensor distribution shifts inherent in optical tactile hardware. In this work, we introduce TouchReason-1M, a large-scale multimodal dataset comprising over 1M synchronized tactile pairs across four distinct sensors, and TouchReason-Bench, a rigorous framework for evaluating tactile perception and visual-tactile conflict resolution. Building upon these, we propose Touch-R1, a tactile reasoning MLLM based on Qwen2.5-VL-7B. Touch-R1 is trained via a tactile-grounded GRPO objective that combines ordinal-aware accuracy, cross-sensor physical consistency, structured-format control, and an input-side tactile grounding objective. Specifically, the tactile-use reward assigns credit only when authentic tactile inputs yield superior correctness relative to counterfactual controls where the tactile stream is removed, shuffled, or noise-masked. On TouchReason-Bench, Touch-R1-7B outperforms Octopi-13B by 18.4% and GPT-4o by 24.7% on average. Its structured reasoning traces reveal emergent behaviors of probing, comparison, and revision, demonstrating that R1-style reasoning can be effectively grounded in physical contact.

Repeated sampling with a verifier is the standard way to allocate test-time compute for code generation, with pass@$K$ as the canonical metric. Yet the standard policy class draws $K$ independent samples from a single answer distribution, so attempts often collapse onto near-duplicate reasoning paths and waste the budget on redundant rollouts. This failure is costly in competitive programming, where many problems admit multiple distinct algorithmic strategies and pass@$K$ requires only one correct attempt. We propose Coordinated Pass@$K$ Policy Optimization (CPPO), which turns pass@$K$ generation into joint exploration over strategies: a planner emits a tuple of $K{=}4$ alternative high-level methods, and a shared solver attempts one solution per method. CPPO trains this joint policy with a multiplicative planner reward, $R_{\mathrm{plan}} = J_ψ\cdot R_{\mathrm{out}}$, assigning credit only to valid strategy tuples that lead to verifier-confirmed pass@$K$ success. Across APPS, CodeContests, and LiveCodeBench-v6, CPPO improves pass@$4$ over direct sampling, planning baselines, planner-only SFT, and pass@$K$-oriented RL under the same $K{=}4$ solver-attempt budget, with statistically significant gains on six of nine model--benchmark cells. The largest single gain is $+0.16$ on Qwen3.5-9B LiveCodeBench-v6 over the strongest baseline, PKPO ($0.588 \rightarrow 0.748$; paired bootstrap, $p < 0.05$).

Group-based reinforcement learning (RL) methods have achieved remarkable success in improving the performance of large language models (LLMs) and have been rapidly extended to agentic tasks. However, their credit assignment relies heavily on coarse-grained trajectory-level attribution according to final outcomes, making it difficult to capture the contribution of individual steps, such as valuable steps obscured within failed trajectories. To uncover latent information and enable more faithful step-level credit assignment, we propose Graph-based Group Policy Optimization (GraphGPO), which first aggregates all rollout trajectories into a unified state-transition graph and then estimates the distance from each state to the task goal using the global information encoded in the graph. Finally, GraphGPO assigns credit to each edge by estimating a graph-based advantage, based on how much the transition reduces the distance to the task goal. In this way, GraphGPO significantly improves training efficiency and achieves state-of-the-art performance across a range of challenging benchmarks.

Long-horizon decision problems with cumulative damage couple locally attractive actions to globally adverse outcomes. We identify two orthogonal failure modes for policy-gradient methods on this class and propose a decomposition that separates them: \emph{completion} (reaching the terminal horizon rather than exiting via an implicit terminal constraint) and \emph{optimality} (matching the dynamic-programming reference given completion). Under PPO with a linear soft penalty, granting horizon access alone reduces the completion rate: the penalty's equilibrium drives the dominant-activity share to zero, while action-space restriction combined with horizon access achieves completion but leaves an optimality gap ($ΔM_{\text{final}} = 0.271$) that we trace to first-phase greedy commitment at the damage origin. We derive four testable predictions and evaluate them in two separately calibrated environments that share the same abstract structure but differ in domain, horizon, activity set, and calibration data: a 49-step bricklayer career and a 20-season NBA power-forward career. All four predictions replicate qualitatively. The horizon-invariance prediction is met at three of four tested horizons, with the exception at $H = 15$ consistent with the $H^*$ boundary ($H^* \in [6, 14]$ under the NBA parameters).

Reinforcement learning (RL) often has a hierarchical structure, where an upper-level (UL) learner selects model parameters and a lower-level (LL) decision-making process responds, naturally leading to a bilevel optimization problem. Most existing bilevel RL methods assume a single-policy LL Markov decision process (MDP), and therefore fail to capture competitive structures arising in applications such as incentive design, where multiple policies interact. We study bilevel optimization problems in which the LL problem is a regularized min-max zero-sum Markov game and the UL objective is optimized through the saddle-point equilibrium induced by the LL game. In this work, we propose penalty-augmented Nikaido-Isoda descent-ascent (PANDA), a penalty-based first-order policy-gradient method based on the Nikaido-Isoda function. By exploiting the min-max game structure, PANDA avoids computing UL hypergradients and does not require second-order information. We prove that PANDA converges to stationary points without convexity assumptions on either the UL or LL objectives. Moreover, PANDA reaches an $ε$-stationary point in $\tilde{\mathcal{O}}(ε^{-1})$ iterations with sample complexity $\tilde{\mathcal{O}}(ε^{-3})$, matching the best-known rates for bilevel RL with single-policy LL MDPs. Experiments demonstrate the superior performance of PANDA over closely related baselines.

We study the sample complexity of policy gradient for log-growth control -- the problem of learning, from observed state transitions, a feedback gain that optimally stabilizes a scalar linear system driven through a multiplicative-noise actuation channel. The objective $J(K) = \mathbb{E}[\log|1+BK|]$ is the top Lyapunov exponent of the closed loop. This problem carries a structural difficulty we call the cusp obstruction: the optimal gain $K^*$ always places the noise singularity $b_{\rm sing}(K) = -1/K$ in the interior of the support. At this singular optimum the policy gradient exists only as a Cauchy principal value, not as a Lebesgue integral, and the natural single-sample gradient estimator has infinite variance. Standard first-order stochastic-optimization analysis is thus inapplicable at the optimum, and merely smoothing the objective does not resolve the difficulty. The obstruction, however, has an exploitable symmetry: the Cauchy kernel is an odd function of the displacement from the moving pole, so pairing each observation with its reflection through the pole cancels the divergent part. This one cancellation simultaneously controls the population curvature, the gradient-estimator variance, and the bias incurred when the noise density is estimated. Combining these bounds with a closed-form single-transition gradient oracle, we prove that projected mini-batch policy gradient, initialized in any compact subset of the stabilizing region, attains total sample complexity $\tilde{O}(1/η)$ when the noise density is known and $\tilde{O}(η^{-(2s+1)/(2s)})$ when it must be estimated, for $C^s$ noise densities with $s \geq 2$.

Reinforcement learning (RL) is the dominant paradigm for post-training large language models. However, in the online, on-policy setting, rollout generation dominates the computational cost of training. Group-based policy optimization methods compute advantages from multiple rollouts per prompt, yet they indiscriminately allocate budget to prompts with collapsed reward distributions, wasting expensive rollouts on negligible learning signals. We demonstrate that group-based updates are most effective in regimes of high reward variance. Since the policy evolves throughout training, prompt informativeness must be estimated online rather than precomputed, but exhaustively evaluating every prompt is computationally prohibitive. We introduce Pilot-Commit, a budget-aware rollout allocation framework for group-based RL post-training. Pilot-Commit decouples prompt evaluation from exploitation: a pilot stage estimates per-prompt informativeness using a fraction of the budget, and the remaining rollouts are allocated to high-leverage prompts while low-signal prompts are skipped. Across multiple math reasoning benchmarks and model scales from 1.5B to 14B parameters, Pilot-Commit matches baseline accuracy with significantly lower sampling costs, reaching target accuracy up to $1.9\times$ faster than GRPO and $4.0\times$ faster than DAPO in cumulative rollouts.

We present the stochastic decoupled policy gradient (SDPG), a lightweight visual reinforcement learning (RL) method that trains diverse visuomotor control policies end-to-end within a few hours on a single NVIDIA RTX 4080 GPU. SDPG estimates policy gradients via random perturbations of trajectory rollouts, requiring orders of magnitude fewer batch-rendered environments and substantially reducing compute and memory overhead. On visual MuJoCo benchmarks, SDPG consistently outperforms baseline methods in training time, memory usage, and rewards. Finally, to support future research, we introduce a suite of realistic visual robotics benchmarks spanning dexterous manipulation, challenging locomotion, and demonstrate effective sim-to-real transfer on physical hardware.

Large-scale search, recommendation, and retrieval-augmented generation (RAG) systems typically employ a two-stage architecture: an early-stage ranker (ESR) generates a candidate set, which is subsequently re-ranked by a late-stage ranker (LSR). While there are many reinforcement learning (RL) methods for training the LSR, end-to-end training of the ESR has proven challenging. In particular, naive application of "vanilla" policy gradient (V-PG) is not scalable for candidate-set sizes relevant for practical use due to exploding variance. This issue arises because V-PG propagates the gradient to the joint probability of the candidate sets, ignoring the contribution of each specific item in the candidate set to the reward. To mitigate this issue, we propose a novel "credit-assigned" policy gradient (CA-PG), which computes gradients with respect to the probability that the target item is chosen in any candidate set, i.e. marginalizing over all candidate sets that contain it. Our theoretical analysis reveals that CA-PG significantly reduces the variance of V-PG by marginalizing over the specific composition of the candidate set, while preserving the ability to learn the correct ranking of items under a reasonably aligned LSR policy. Experiments on both synthetic and real-world data demonstrate that CA-PG improves the convergence speed and training stability for ESRs utilizing the canonical Plackett-Luce model, especially when the candidate-set size is large.

Model-based reinforcement learning (RL) can be effectively supported at scale through the use of world models. However, in practice, scaling such approaches remains fundamentally limited. A commonly recognized challenge is model bias and error compounding, which degrade long-horizon predictions. Beyond these issues, we identify a more critical yet underexplored bottleneck: a structural misalignment between search and value learning in existing world model approaches. In particular, policy improvement often relies on value functions induced by a separate, non-search policy, resulting in training inconsistency and ultimately suboptimal learning. To address this limitation, we propose Model-Based Diffusion Policy Optimization (MBDPO) in world models, a framework that unifies search and policy optimization through diffusion policy representations, thereby unlocking the potential of world models for scalable policy learning. Instead of constructing an explicit planner over a learned world model, we reformulate policy optimization as a diffusion process over searched trajectories in latent world models. In this view, we extract an implicit energy function from the collected dataset that anchors the policy, enabling MBDPO to refine the score field for policy optimization while mitigating misalignment. We evaluate MBDPO across a wide range of settings, including multi-task offline pretraining, online learning, and offline-to-online fine-tuning. In the offline regime, we further investigate its scaling behavior by pretraining on large-scale datasets, observing consistent and monotonic performance gains with increasing model capacity.

Wasserstein policy gradient (WPG) is a policy optimization method for reinforcement learning (RL) that exploits the optimal-transport geometry of action distributions. For the entropy-regularized RL objective, WPG evolves each state-conditional policy by transporting it along the action gradient of the soft Q-function together with a Langevin-type diffusion. Despite its appeal for continuous-control problems, its global convergence properties remain poorly understood. Standard Langevin analyses do not directly apply, because the RL objective depends on the policy through the Bellman recursion rather than through a static convex functional, and the Langevin drift is determined by the soft Q-function, whose regularity must be controlled along the policy iterates. In this paper, we develop a global convergence theory for WPG by exploiting the Bellman structure of entropy-regularized RL. We show that the role usually played by convexity can be replaced by a Bellman-based argument: the soft Bellman residual admits a statewise KL representation with respect to a Gibbs policy; Bellman contraction relates this residual to the global optimality gap; and a Bellman resolvent identity connects value improvement to relative Fisher information. Combined with a uniform log-Sobolev inequality (LSI) for the evolving Gibbs family, these ingredients yield a distributional Polyak--Łojasiewicz condition. We further establish the regularity and uniform bounds needed to control the discretization error, thereby obtaining geometric contraction up to a discretization bias. Conceptually, our analysis shows that although entropy-regularized RL is not convex in the usual flat sense, the Bellman recursion induces a favorable Polyak--Lojasiewicz-type (PL) geometry that supports global convergence of WPG.

This paper applies machine learning to the difficult and important task of version control merging. (1) We constructed a dataset, Merge-Bench, of 7938 real-world merge conflict hunks from 1439 GitHub repositories. The ground truth is the merge resolution that developers committed to the repository. Our dataset construction methodology is scalable to arbitrary amounts of data since no manual labeling is required. (2) We trained a model, LLMergeJ, to resolve merge conflicts in Java programs. Our approach uses Group Relative Policy Optimization (GRPO), an online reinforcement learning method, to train a Large Language Model (LLM). (3) We performed two evaluations of the performance of LLMs on resolving merge conflicts. On Java programs, LLMergeJ with 14B parameters outperforms 3 commercial LLMs, trailing only Gemini 2.5 Pro. Across 11 programming languages, commercial LLM performance is largely stable from language to language. The best models correctly resolve less than 60% of merge conflicts.