Reinforcement_Learning
-
Recursive Reward Improvement via Adversarial Meta-Prompt Evolution
2026-10-15 00:31:18Wei Dai, Qianchu Liu, Guanghui Qin, Xing Han, Tristan Naumann, Paul Pu Liang, Sheng ZhangAbstract
Recursive Self-Improvement (RSI) has gained wide attention for its ability to work with limited data. In a proposer-solver loop, a proposer model writes training tasks and a solver model learns to solve them through reinforcement learning. Such loops have succeeded in mathematics and coding, where every answer can be checked automatically. Open-ended expert tasks, like clinical report generations, have no answer key, which makes simple question-answer generation not viable. Here the proposer must also write a rubric that lists the content an answer should include and the errors it should avoid. However, the unique challenge behind model-written rubric is the comprehensiveness of it. For naively generated rubric, solvers tend to exploit it, naming the expected topics without the substance behind them, which corrupts the learning process. To solve this problem, we introduce RRI, Recursive Reward Improvement, which uses a reward hacking adversary to supervise the reward. Before the solver trains on a generated task, an adversary writes an answer that games its rubric while being as useless as possible, the same model writes an honest answer without seeing the rubric, and up to two repair steps revise the rubric to make the honest answer win. The exploits found also rewrite how future tasks and rubrics are generated. On HealthBench Professional, RRI improves the adjusted score over the Generation Only self-improvement baseline by 10.9 points at 9B and 12.4 at 27B, and by 10.3 points without any external model. In particular, the gains are largest on difficult and writing tasks. We show that the 27B model trained on the HealthBench Professional description improves by 6.6 points on PRBench Hard and 8.0 on ProfBench without seeing a task from either. We further analyze the exploitable points of model-generated question-rubric pairs to facilitate future research on open-ended self-play.
Abstract (translated)
URL
https://arxiv.org/abs/2610.21040
PDF
-
Learning at the Right Pace: Adaptive Data Scheduling Improves LLM Reinforcement Learning
2026-10-14 23:58:09Zicheng Xu, Ruixuan Zhang, Yu-Neng Chuang, Xiuyi Lou, Hoang Anh Duy Le, Oren Gal, Alexander S. Szalay, Zhaozhuo Xu, Guanchu Wang, Vladimir BravermanAbstract
Large Language Models (LLMs) achieve remarkable reasoning capabilities through reinforcement learning (RL) post-training. However, existing RL post-training commonly relies on uniform data sampling, which ignores the semantic structure of the training data and the evolving capability of the training policy. To address these limitations, we propose Adaptive Data Scheduling (ADS), a dual-level data scheduling framework that combines adaptive sampling across semantic clusters with policy-boundary sample selection. At the cluster level, ADS organizes samples according to semantic patterns and adapts the inter-cluster sampling distribution to prioritize clusters within the policy's current reach. At the sample level, ADS tracks prompts near the policy's capability boundary within each cluster to obtain informative relative advantages. Experimental results across three LLMs and seven reasoning benchmarks demonstrate that ADS improves average accuracy by 5.5% over Group Relative Policy Optimization (GRPO). Notably, ADS consistently improves RL methods with different objective designs, highlighting its potential as a general data scheduling strategy. The source code is available at: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2606.22305
PDF
-
Deep Reinforcement Learning for HPC Job Scheduling: A Statistical Evaluation
2026-10-14 23:02:02Justin M. Cheney, Clement N. NyirendaAbstract
Deep Reinforcement Learning (DRL) offers learned job selection for High-Performance Computing (HPC), but performance against heuristics alone does not establish value beyond action masking. Six DRL treatments are compared: Proximal Policy Optimization (PPO), Advantage Actor-Critic (A2C), and Deep Q-Network (DQN), each with and without masking. Three heuristics and a seed-paired masked uniform random control provide baselines across ten seeds on two real Slurm (Simple Linux Utility for Resource Management) workload traces, one homogeneous and one heterogeneous, under a common non-parametric framework. All masked treatments complete evaluation, whereas unmasked treatments are truncated. On the homogeneous trace, no masked DRL treatment significantly improves average waiting time or average and maximum bounded slowdown over the control. On the heterogeneous trace, MaskablePPO improves all three metrics relative to the control. Its average waiting time is equivalent to that of the three heuristics within a 10% margin, and its maximum bounded slowdown is 22% lower than that of Last-Come, First-Served (LCFS). MaskablePPO has the leading average waiting-time rank among complete treatments on both traces, but Nemenyi comparisons detect no significant difference from MaskableA2C. Among masked variants, MaskablePPO has the lowest recorded mean training time, while MaskableDQN has the lowest mean decision latency. These findings distinguish completion from baseline-specific scheduling gains and support including a masked random control when assessing the value of learned selection.
Abstract (translated)
URL
https://arxiv.org/abs/2610.21008
PDF
-
How to Train Your Little Dragon: Teaching a Small Language Model to Construct and Refine GPU Optimization Scripts
2026-10-14 21:49:24Annabelle Sujun Tang, Wei Tang, Songyao Jiang, Tianhui Li, Shengzhe Xu, Jiin Woo, Kaan Ozkara, Youngsuk Park, Zhen Jia, Hadi EsmaeilzadehAbstract
Compiler transformations interact, making GPU kernel optimization a search for combinations that preserve the computation while improving performance. We introduce HOW TO TRAIN YOUR LITTLE DRAGON, a training method that teaches a small language model both to construct complete optimization scripts and to improve existing ones. Instead of ranking candidates or generating kernel code, our model generates optimization scripts. We devise a compact, lossless representation whose deterministic decoding reconstructs the exact compiler trace, preserving its transformation parameters and dependencies. We use a compiler framework TVM MetaSchedule to build training data on 54,116 distinct workloads, and construct 108,594 examples in two training sets: 60,236 teach the model to generate an optimized script from an unoptimized kernel, while 48,358 teach it to produce a better script given the kernel and a somewhat optimized script. Supervised fine-tuning teaches the model to construct and improve optimization scripts from these search-derived examples. We then add reinforcement learning with a shaped reward that combines correctness with measured performance, awarding a latency-based bonus only to numerically correct kernels. We apply this training method to a model with 8 billion parameters. After supervised fine-tuning and reinforcement learning, the model achieves per-attempt correctness of 91.27% for construction and 92.01% for refinement, while reinforcement learning produces faster kernels than supervised training on 88 of 102 construction tasks and 145 of 170 refinement prompts. With only 8B parameters, our LITTLE DRAGON achieves higher per-attempt correctness and better kernel performance than six frontier LLMs: GPT-6 Astra, Opus 5, Kimi-K3, and GLM-5.3 among them, each with hundreds of billions to trillions of parameters, for which task-specific training is unavailable or expensive.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20980
PDF
-
SOMA: Stack-of-Objects Manipulation using a Quadruped Robot
2026-10-14 21:43:14Stergios E. Bachoumas, Panagiotis ArtemiadisAbstract
Executing heavy-duty physical manipulation in unpredictable environments remains a major challenge for quadrupedal robots lacking dedicated manipulators. This paper presents SOMA, a hierarchical reinforcement learning framework for whole-body non-prehensile manipulation of multi-object stacks using a quadruped robot. Unlike traditional methods restricted to two-dimensional planar pushing, SOMA integrates explicit robot base-height control, unlocking full three-dimensional pushing strategies. This capability enables the quadruped to manipulate complex payload stacks as a unified system while preventing relative sliding or tipping of upper layers. The architecture factorizes control into a high-level manipulation policy using a Simplified Structured State Space Sequence (S5) encoder to reason over long-horizon temporal history and predict bounded target tracking commands via Beta distributions, and a low-level locomotion policy that dynamically adapts leg stance configurations. In extensive simulation benchmarks, SOMA achieves a $90.3\%$ success rate for single-object reposing and up to $90.1\%$ for dynamic two-box stacks under stringent error margins ($0.05\text{ m}$ position error, $1^\circ$ orientation error). Finally, we validate zero-shot sim-to-real transfer on a Unitree Go1 quadruped robot across diverse payload masses and geometries, demonstrating active base-height adjustments to maintain traction and balance during heavy pushing tasks.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20968
PDF
-
MAI-Gym: An Open All-in-One Platform for GUI Agent Evaluation and RL Training across Mobile and Computer Environments
2026-10-14 21:06:01Panrong Tong, Quyu Kong, Pengxiang Li, Tianyu Xia, Hanzhang Zhou, Lei Wang, Chenglin Cai, Long Li, Long Chen, Jianan Zhang, Zhengwen Zeng, Xu Zhang, Gongjie Zhang, Yue Wang, Steven HoiAbstract
Online reinforcement learning (RL) with verifiable rewards has driven rapid progress in mathematics, coding, and tool use, but scaling this success to GUI agents remains challenging due to the high cost of interactive environments and the scarcity of realistic training tasks. We introduce MAI-Gym, an open all-in-one platform that unifies GUI agent evaluation, data collection, and online RL training across mobile, computer, and web environments. Through a shared rollout pipeline, MAI-Gym ensures consistent agent behavior across evaluation and training, supports ten established benchmarks, and scales to 20,000 parallel environment instances. Beyond scalable infrastructure, MAI-Gym enables automated synthesis of diverse GUI applications and challenging tasks, including resettable mocked apps and coherent cross-app worlds with execution-validated task verifiers. Using MAI-Gym, we train MAI-Gym-9B, a unified 9B agent for mobile and computer use, achieving state-of-the-art performance among open GUI agents under 10B parameters: 78.4% on MobileWorld and 76.2% on OSWorld-Verified. Importantly, RL in a mocked train-ticketing app transfers to its real-world counterpart, improving task success on a phys- ical phone from 28.8% to 48.1%. We release the framework, synthesis skills, mocked apps, 2,000 verified tasks, and the model checkpoint to support repro- ducible research and further advances in scalable GUI agent training.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20939
PDF
-
Wrong Votes, Right Gradients: Why Majority-Vote Rewards Track Verifiable RL
2026-10-14 20:25:42Dylan P. Wilson, Chengsong Huang, Jiaxin HuangAbstract
Reinforcement learning with self-generated labels, such as majority vote over a model's own samples, often recovers most of the gain of training with verifiable rewards. This is puzzling from the usual prompt-level view, in which majority vote only returns a model's modal answer and cannot add information about correctness. We argue that the relevant unit is the batch-level update. Because parameters are shared, a prompt's own gradient accounts for less than 1\% of the change in that prompt's output at practical batch sizes, so a wrong label matters only to the extent that label errors share a direction across prompts. We decompose the majority-vote update into shrinkage and misdirection relative to the gold update, and show that misdirection vanishes with batch size whenever errors are incoherent across prompts, however often the labels are wrong. Measuring per-response gradients on five models, we find that majority-vote errors are as incoherent as a random-reward null, that the majority-vote update agrees with the gold update as closely as two independent gold updates agree with each other, and that its noise-corrected cosine with the gold update is 1 within error. Randomly flipping up to 30\% of gold labels leaves the update direction intact, while the same fraction flipped by a shared rule destroys it, and rewards that favor a kind of response on every prompt, such as a learned reward model, fall measurably short. The equivalence holds at every checkpoint we examine but is local: gold-reward training still ends ahead on held-out majority accuracy. Error coherence, not error rate, determines when self-generated labels can stand in for verification.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20898
PDF
-
Race to Recovery: Designing Advanced Safety Systems for Autonomous Driving via Learning Autonomous Racing
2026-10-14 20:04:54Teodor Ilie, Kleber Cabral, Sidney GivigiAbstract
Vehicles that lose control at the handling limit through icing, hydroplaning, or an abrupt evasive maneuver fall outside the operating envelope of production safety systems such as electronic stability control, which focus on preventing instability rather than recovering from it. Expert drivers regain authority in these conditions by drifting. Existing autonomous recovery approaches are validated on idealized scenarios, define "recovered" states empirically and inconsistently, or rely on offline-planned trajectories rather than acting online. We address these gaps by first training a racing policy with reinforcement learning that learns drifting as a time-optimal maneuver, then adapting it to recovery through continued training with a minimum-time reward. We define recovery relative to a phase-plane-grounded stable equilibrium and introduce Gym-Khana for controller training and evaluation. Across 294 initial loss-of-control states, the transferred policy improves recovery by 186.8% over a Stanley path tracker and outperforms both a nonlinear model predictive controller and a recovery-only policy trained from scratch, while also recovering faster on states the baseline can already handle. An ablation shows that racing pretraining, rather than broader initial-state exposure alone, drives this gain. Deployed on a physical 1:10-scale F1TENTH vehicle using system identification, actuator-lag modelling, and domain randomization, the transferred policy outperforms Stanley on every tested loss-of-control state, including recovering 50% of the time from a hard state the baseline never recovers from.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20873
PDF
-
Gibbs-Gap Policy Optimization for Off-Policy RL with Verifiable Rewards
2026-10-14 18:59:29Xuanchang Zhang, Tong ZhangAbstract
Reinforcement learning with verifiable rewards often trains language models on responses generated by a different, possibly stale behavior policy. Correcting this mismatch with learner-to-behavior importance weights can introduce high variance for long responses. We introduce Gibbs-Gap Policy Optimization (GGPO), which minimizes a log-mean-exp contrast of KL-adjusted rewards among responses to the same prompt, without learner-to-behavior importance weights. We derive GGPO from KL-regularized policy optimization: on policy, its population loss is exactly the KL-regularized optimality gap, and under suitable coverage the learner's single-response expected reward is at least the behavior policy's best-of-$K$ reward, minus terms for the raw finite-group loss, regularization, and finite training data. We characterize reference refresh through exact Gibbs-target composition and a finite-data final-policy reward bound under local reference control and tokenwise behavior mismatch. This refresh is empirically important: in an ablation, fixed-reference GGPO scores $32.96$ on the five-benchmark problem-weighted average, versus $46.58$ with the default refresh schedule. Experiments on mathematical reasoning further show that GGPO outperforms GRPO and AGRO under training-inference mismatch and on fixed offline data, while maintaining its performance as controlled policy staleness grows.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20814
PDF
-
Blame Errant Words, Not the Canvas: Defect-Aware Spatial Credit Assignment for Visual Text Generation
2026-10-14 18:38:58Sujie Hu, Huaisong Zhang, Pujing Yang, Qian Wang, Xiaomin Li, Xinhang Leng, Chen Li, Jing Lyu, Xiu Li, Ling PanAbstract
Recent reinforcement learning approaches have improved visual text generation, yet generated images still suffer from incorrect wording and malformed glyphs. We identify two coupled limitations: recognition-based rewards often overlook visible glyph defects, while whole-image scalar feedback obscures the uneven quality of individual text regions. These limitations motivate our central principle: $\textit{Blame errant words, not the canvas}$, to effectively preserve local quality distinctions. First, we introduce $\mathbf{TextDART}$, a defect-aware region-targeted reward model that localizes individual text regions and independently evaluates their semantic correctness and glyph quality, providing fine-grained, region-precise feedback. Building on TextDART, we propose $\mathbf{SCANNER}$, which turns these regional judgments into spatially differentiated policy feedback. It normalizes rewards for each text target across sampled images and casts the resulting regional advantages onto the corresponding latent tokens. Extensive experiments on existing benchmarks and our newly established $\textbf{PolyText-Bench}$ demonstrate that our approach achieves state-of-the-art performance in text accuracy and glyph fidelity. Project page: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2610.20794
PDF
-
EvolvingBrain: An Adaptive Skill System for Multi-modal Agents Self-Evolution in Open Worlds
2026-10-14 18:30:07Runxuan Liu, Jingqi Shi, Baoxin Wang, Honghong Zhao, Haoyuan Xu, Yueqing Sun, Jiafeng Liang, Tao He, Dayong Wu, Qi Gu, Shijin Wang, Ming Liu, Bing QinAbstract
Multi-modal agents have made significant progress in interacting with open worlds. However, their ability to self-evolve remains limited. Supervised fine-tuning enables imitation but not active experience organization, while outcome-based reinforcement learning promotes exploration yet underuses intermediate trajectory information. Therefore, we propose EvolvingBrain, which uses an adaptive skill system as the brain for agent self-evolution, consisting of two following components. Stratified Skill Self-Evolution: extracts Procedural Patterns and Problematic Patterns from model-specific trajectories, summarizes experience by generating skills, and evolves through multiple interactions with the open world, ultimately forming a Stratified Skill Architecture consisting of Domain, Plan, and Action/Trap Skills, which provides the agent with a multi-granularity knowledge repository and a novel evolution approach. State-Aware Adaptive Execution: summarizes long-horizon interactions into Structured Abstractive Memory and uses it to guide the State-Aware Skill Fusion Strategy, enabling the agent to utilize knowledge according to its current state. Experiments on VAB-Minecraft and DiscoveryWorld demonstrate that EvolvingBrain improves success rates on open-world tasks and exhibits generalization capabilities. Our code is provided in the supplementary materials and will be publicly released in the camera-ready version.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20786
PDF
-
ReasonUni: Reasoning-Augmented Reconstruction Advances Unified Multimodal Models
2026-10-14 18:27:53Feize Wu, Zhiyuan Yan, Yuhuan Hong, Yuwei Niu, Zongjian Li, Zheyuan Liu, Bin Lin, Jinbo Yan, Jie Qin, Xudong Mao, Limeng Qiao, Guanglu Wan, Li YuanAbstract
Image-to-text (I2T) understanding and text-to-image (T2I) generation are dual tasks in unified multimodal models (UMMs). Reconstructive post-training couples these tasks, but existing reconstruction-based methods remain perception-centric and provide limited supervision for reasoning about structure, intent, and transformation. We propose ReasonUni, a reasoning-augmented reconstructive post-training framework that extends this coupling beyond perception. Given an original image and an editing instruction, the UMM first reasons about what and how to edit and describes the target state, then generates the corresponding target image conditioned on this understanding output. To jointly optimize both modules, we introduce TreeGRPO, a tree-structured reinforcement learning method that samples multiple reasoning paths and candidate images per path, then aggregates visual rewards to update both modules under a unified objective. Experiments across multiple benchmarks demonstrate consistent gains in both understanding and generation, showing that reasoning-augmented reconstruction yields genuine mutual gains for UMMs.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20781
PDF
-
Modeling and Understanding Neural Weights of Humanoid Policies with Diffusion
2026-10-14 17:59:52Huixuan Cheng, Bike Zhang, Shankar Sastry, Koushil SreenathAbstract
Recent advances in robot learning have driven rapid progress in humanoid control, enabling increasingly agile behaviors in the real world. However, humanoid control networks are typically obtained through conventional training procedures, while comparatively little attention has been given to directly modeling the distribution of their learned weights and generating functional networks in weight space. In this work, we investigate whether high-performing humanoid control networks can be generatively modeled directly through their weights. We adopt a latent diffusion framework that couples a linear autoencoder with a Diffusion Transformer. The autoencoder compresses actor network parameters into a compact latent space, and the Diffusion Transformer models the distribution of these latents. Trained on networks collected from reinforcement learning for the Unitree G1 robot, our approach generates deployable networks that perform various motions in the real world, including walking, dancing, and spin kick. Beyond generation, we further study the learned weight space to better understand the structure captured by these humanoid control networks. We identify a structural principle in the first-layer weights that consistently reflects the robot's kinematic prior. Notably, this kinematic prior emerges under both Reinforcement Learning (RL) and Imitation Learning (IL) training, suggesting that it is not specific to a particular optimization process but may instead reflect common knowledge associated with humanoid embodiment. Together, these results demonstrate the feasibility of generative modeling in humanoid network weight space, while providing initial insights into the information encoded by their learned weights. Additional results are available on our project website: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2610.20606
PDF
-
Player-1: Learning to Play from Synthetic Experience
2026-10-14 17:57:12Mingyu Ouyang, Yiran Wang, Siyuan Hu, Xingyilang Yin, Huiqiang Sun, Guangzhi Wang, Junfu Pu, Cheng Cheng, Honghao Chen, Ying Shan, Mike Zheng Shou, Hwee Tou NgAbstract
Computer-use agents have demonstrated strong capabilities in interpreting screens and executing keyboard and mouse actions. However, games, with their visually rich worlds and diverse mechanics, remain challenging and offer a demanding next step toward general digital control. Current game-agent learning pipelines often rely on large-scale demonstrations with limited coverage of dynamic game states, while scalable environments for interactive training remain difficult to obtain. We introduce Player-1, a post-training framework for learning game control beyond demonstrations through synthetic experience. We extend the computer-use action space with relative pointer motion, sustained inputs, and composite actions. For supervised initialization, we fine-tune computer-use models on 6,044 curated gameplay trajectories that pair recorded actions with task instructions and visual reasoning. We construct Player-1-Gym, sixteen synthetic 2D and 3D environments designed around control demands shared across games. Programmatically verifiable rewards and randomized observations and control dynamics support multi-environment reinforcement learning with PPO. The resulting Player-1-9B and Player-1-35B-A3B improve MineDojo-300 success over their SFT policies by 27.8 and 17.7 percentage points, respectively. Player-1-9B reaches 38.33% success compared with 12.63% for the strongest evaluated baseline, with further gains over SFT on ViZDoom-10 and GameWorld. Further analyses show that broader synthetic environment coverage improves transfer, while the use of composite actions that combine multiple keyboard and mouse inputs increases during RL. We demonstrate that learning from synthetic experience substantially improves game control beyond supervised initialization, with gains that generalize beyond the RL training suite.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20584
PDF
-
PrismaCoder: Pushing the Boundaries of Open Multimodal Code Intelligence through Decomposed Rubric Synthesis
2026-10-14 17:32:54Qiushi Sun, Jingyang Gong, Yang Liu, Qiaosheng Chen, Zhangyue Yin, Zixian Huang, Hanxu Hu, Jing Xiong, Fangzhi Xu, Haochen Tan, Haoli Bai, Lifeng Shang, Xiaoguang Li, Ben Kao, Qipeng Guo, Lingpeng Kong, Jiahui GaoAbstract
Multimodal code intelligence connects the symbolic world of programs with the visual world their execution creates. Interest in this capability is burgeoning, as models are increasingly asked to produce charts, webpages, animations, and semi-structured documents. Recently, proprietary models have pushed far ahead, producing interactive artifacts even from a text description or a screenshot alone. Yet their data and training remain closed, and open progress is held back by the quantity and quality of available data and by the pipelines and methods for producing it. Quality is the harder of the two, since a sample must be correct as a program, well built, and above all visually right once rendered. No existing signal covers all three. Execution often returns just a pass or fail, code similarity does not map to the visual output, and task-specific verifiers do not generalize. To push open progress further, we start from the data side with PrismaForge, a hierarchical data pipeline centered on a synthesized rubric that decomposes these factors into checkable criteria. The same rubric serves as both filter and reward: it curates the supervised corpus, then drives reinforcement learning on visual-coding tasks. Leveraging PrismaForge for data synthesis and filtering, we build PrismaCoder-2M, the most diverse open dataset for multimodal code intelligence and the largest to date. On the model side, we train PrismaCoder-9B and PrismaCoder-35B on this data and release both. They lead the open state of the art by a wide margin and rival leading proprietary systems. Our analyses further show that the proposed rubric-based pipeline improves multimodal coding performance across model backbones. We release our code and data at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20538
PDF
-
Asynchrony Is Nearly Free for Evolution Strategies on Long-Horizon LLM Agentic Tasks
2026-10-14 16:48:33William Hoy, Jingxuan Fan, Nurcin Celik, Xu PanAbstract
Synchronous policy updates can limit the efficiency of post-training for long-horizon LLM agents. When trajectories span multiple interaction turns, completion times can vary substantially, and synchronous update barriers leave faster workers idling, wasting compute resources. Asynchronous reinforcement learning addresses this inefficiency but introduces policy lag and off-policy optimization. Evolution strategies (ES) are a backpropagation-free alternative to gradient-based training, yet require many perturbed-policy evaluations and have remained largely synchronous in current LLM post-training. We introduce bounded-staleness asynchronous ES and investigate its training behavior under stale fitness measurements. On WebShop, we systematically vary the lag and fraction of stale fitness measurements and additionally introduce two stress tests to probe extreme staleness cases. Natural asynchrony closely tracks synchronous ES, and degradation in evaluation accuracy remains small even under extreme staleness, including when all fitness measurements are stale. These findings suggest that ES can tolerate substantial policy staleness without explicit off-policy correction, supporting its use in asynchronous post-training for long-horizon LLM agents.
Abstract (translated)
URL
https://arxiv.org/abs/2610.04196
PDF
-
Safe Reinforcement Learning with Receding-Horizon Safety Critic and Differentiable Action Shields
2026-10-14 14:25:46Siwei Ju, Jan Peters, Oleg ArenzAbstract
Despite the Markov assumption, safety failures in robot reinforcement learning rarely occur instantaneously; they typically emerge over several steps as the policy enters unfamiliar states under disturbances and domain shift. We address this limitation by learning a receding-horizon safety critic that predicts whether a state will remain non-terminating over the next receding horizon, enabling anticipatory detection of impending locomotion failures. The critic is trained from rollout-derived horizon labels with a composite objective that combines horizon classification, control-barrier-function-inspired temporal consistency, and smoothness regularization, and we aggregate an ensemble of critics with a conservative Conditional Value at Risk estimate to account for uncertainty. Building on the learned critic, we derive a closed-form, differentiable CBF-style safety filter that minimally modifies the policy action to improve predicted safety, and integrate it into PPO with an anticipation loss to reduce policy-filter mismatch during training. In simulated quadruped locomotion, our method improves robustness under action noise, external perturbations, and terrain shifts, reducing catastrophic failures while maintaining competitive returns.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20255
PDF
-
MedSonor: Learning to Use Clinical Audio in Medical LLMs
2026-10-14 14:04:43Geyang Yu, Junying Chen, Benyou WangAbstract
Clinical audio conveys medical information that transcripts may not preserve, but using it in medical responses requires interpreting it alongside clinical requests and dialogue context. Existing medical audio supervision largely targets classification or recognition, while general audio-language models lack specialized medical knowledge and direct feedback for open-ended responses. We present MedSonor, a medical dialogue model integrating medical knowledge, spoken dialogue, and clinical audio understanding. We pair authentic medical audio with clinical requests and combine these audio-integrated dialogues with collected medical audio and synthesized spoken dialogues for supervised fine-tuning, followed by two stages of rubric-guided reinforcement learning. We introduce MedVoice-Dial and MedVoice?AID to evaluate responses within supplied multi-turn medical contexts and to requests involving clinical audio, respectively. Across seven medical benchmarks, MedSonor outperforms the evaluated general-purpose open-weight audio-language baselines and ranks highest in aggregate physician preference evaluation, while broadly retaining general speech capabilities.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22405
PDF
-
Noise-Aware Lombard Speech Synthesis via Flow-GRPO for Intelligibility Optimization
2026-10-14 13:29:29Zhaoyang Wang, Dingding Yao, Chengzhong Wang, Junfeng LiAbstract
Voice-based conversational agents are increasingly deployed in real-world settings, where environmental noise often degrades the interactive experience. Therefore, robust acoustic adaptability is essential to maintain natural and intelligible human-computer communication. However, most existing systems remain acoustically insensitive and fail to emulate the human Lombard effect, in which speakers involuntarily adjust their vocal production to preserve intelligibility in the presence of noise. As a result, their communicative reliability degrades under adverse acoustic conditions. To remain intelligible in noisy environments, conversational agents require a mechanism to emulate noise-induced adaptive behavior during synthesis. We propose NIL-TTS, a noise-induced Lombard TTS framework. We integrate a feature-wise linear modulation (FiLM) mechanism into a flow-matching backbone for dynamic vocal adaptation. A two-stage training paradigm is utilized: initial supervised fine-tuning on paired Lombard corpora, followed by Flow-GRPO post-training to optimize for intelligibility. Evaluations confirm that NIL-TTS autonomously adjusts its output in response to varying noise conditions, exhibiting key correlates of the human Lombard effect. Driven by these adaptive modulations, the synthesized speech achieves significant intelligibility improvements that scale proportionally with noise intensity. Ultimately, this work presents a flow-matching and reinforcement learning-based advancement within the adaptive Lombard speech synthesis paradigm for challenging real-world scenarios.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20181
PDF
-
RM-OPD: On-Policy Distillation for Generative Reward Models
2026-10-14 13:20:55Qinglin Su, Licheng Pan, Zhen Fang, Jiakang Wang, Qingnan Ren, Yunsheng Lu, Yan Gao, Zhixuan ChuAbstract
Reinforcement Learning on Mixed-domain preference data (Mix-RL) has become the mainstream approach for training general-purpose Generative Reward Models (GRMs). However, Mix-RL suffers from two critical bottlenecks: the reward sparsity induced by outcome rewards, and the cross-domain interference arising from jointly optimizing heterogeneous capabilities, which together give rise to imbalanced capability acquisition and pervasive reward hacking. Inspired by the success of On-Policy Distillation (OPD) in multi-domain capability integration, we introduce RM-OPD (Reward Modeling via On-Policy Distillation), a two-stage framework that separates specialist acquisition from capability integration. First, we cultivate domain-specialized teacher models via single-domain GRPO fine-tuning, allowing each expert to reach its performance ceiling in isolation; then we consolidate these teachers into one student model through (i) routed on-policy distillation, which provides dense signed token-level supervision on student-generated critiques; (ii) domain contribution reweighting, which reweights token guidance toward target domain shares, preventing specific domains from dominating optimization; and (iii) optional criteria context distillation, which conditions teachers on domain-specific evaluation criteria to sharpen token-level supervision. Across multiple RM benchmarks and two base models, RM-OPD consistently outperforms vanilla Mix-RL and other strong baselines, achieving overall superior performance and exhibiting an emergent "teacher-surpassing" effect. These results establish RM-OPD as a promising paradigm for building generalist GRMs.
Abstract (translated)
URL
https://arxiv.org/abs/2610.20174
PDF
