The paper presents a latent diffusion framework that models the distribution of neural network weights for humanoid control policies. By compressing actor network parameters with a linear autoencoder and modeling the latent space with a Diffusion Transformer, the authors generate deployable networks that perform real‑world motions such as walking, dancing, and spin kicks on the Unitree G1 robot. The study also reveals a structural principle in the first‑layer weights that reflects the robot’s kinematic prior, appearing consistently across both reinforcement and imitation learning training methods.
By Huixuan Cheng, Bike Zhang, Shankar Sastry, Koushil Sreenath
The study evaluates scheduling strategies for echocardiographic exams in a pediatric echo lab, comparing on-the-fly allocation with reservation-based approaches using real operational data from Stanford’s Lucile Packard Children’s Hospital. On-the-fly allocation consistently reduces patient waiting times across all tested reservation configurations. A reinforcement‑learning model (Dueling Double DQN) was trained to approximate an optimal on-the-fly policy but did not outperform the best rule‑based on-the-fly strategy.
By Bozhi Sun, Seda Tierney, Jeffrey A. Feinstein, Frederick Damen, Alison L. Marsden, Daniele E. Schiavazzi
This paper tackles multi‑objective imitation learning, aiming to recover policies on the Pareto front from demonstrations by multiple Pareto‑optimal experts in a Multi‑Objective Markov Decision Process. The authors propose Multi‑Output Augmented Behavioral Cloning (MA‑BC), which partitions divergent expert data and pools non‑conflicting state‑action pairs, proving that MA‑BC converges to Pareto‑optimal policies faster than independent‑dataset learners and is minimax optimal. Experiments on diverse discrete environments and a continuous Linear Quadratic Regulator confirm the algorithm’s effectiveness.
By Ziyad Sheebaelhamd, Luca Viano, Volkan Cevher, Claire Vernade
The paper surveys 58 studies on verifier errors in reinforcement learning with verifiable rewards (RLVR), examining how such errors affect training and which mitigations are effective. It finds that the type of error—rather than its frequency—has a larger impact on learning dynamics, with random errors mainly slowing progress and false positives potentially causing learning plateaus or collapse. The authors compile findings into comparison tables, reconcile conflicting reports, and propose a reporting protocol to standardize future research.
By Savyasachi Gupta, Aniruddha Srinivas Joshi
DriveWLA is a unified World‑Language‑Action model that integrates semantic understanding, future scene dynamics, and trajectory generation for autonomous driving. It employs a Mixture‑of‑Transformers architecture with three experts—World, Language, and Action—connected by Cross‑Expert Attention to align semantic and dynamic representations. After reinforcement learning refinement, DriveWLA outperforms existing VLM‑based and world‑model‑based planners on NAVSIM benchmarks and generalizes well to other datasets without fine‑tuning.
By Shuzheng Wang, Haoxuan Li, Shengzhen Li, Chenxi Wu, Yajing Sun, Lu Hou, Wenyu Liu, Xinggang Wang
MERIT (Multimodal Evidence Reasoning over Image Tables) is a three‑stage framework designed to enable language models to reason over table images, which preserve visual cues and structural details that serialized textual tables often lose. The approach first adapts a table‑specific encoder through reconstruction, localization, and color‑conditioned supervision to address visual evidence acquisition. It then fine‑tunes the model with supervised evidence selection and reasoning, followed by reinforcement learning that rewards evidence correctness, calculation, final answers, and output format, thereby improving retrieval and computational accuracy in table question answering and fact verification tasks.
By Haoyang Jiang, Zhuotong Fan, Mingyang Yi, Ju Fan, Lanqing Hu, Xiao Zan
The paper introduces ProAS, a Progress‑Aware Advantage Shaping framework designed to improve critic‑free reinforcement learning for large language model agents. ProAS uses a dynamic Agentic Reference Graph to provide fine‑grained credit assignment through Masking Unreasonable Incentives and Asymmetric Penalty Redistribution, avoiding the uniform advantage broadcast that can reinforce invalid actions. Experiments on Qwen2.5‑1.5B‑Instruct and Qwen2.5‑7B‑Instruct show that ProAS raises baseline success rates to 91.9% and 94.3% on ALFWorld and 66.7% and 70.8% on WebShop, while speeding up training convergence up to 2.44× and remaining robust to imperfect reference graphs.
By Bowen Ding, Mingda Wan, Yuhan Chen, Jiyao Yuan, Qi Zhu, Futing Wang, Heyuan Deng, Fei Mi, Lifeng Shang, Tao Lin
The paper investigates reinforcement learning with verifiable rewards (RLVR), where costly model-generated responses encourage larger rollout batches and a square‑root learning‑rate scaling. Using a two‑clock framework, the authors show that while the square‑root rule preserves first‑order drift per sample, larger batches suffer from curvature losses, leading to a critical batch size beyond which sample efficiency degrades. Experiments on Qwen2.5‑Math‑1.5B confirm the predicted progress gap and failure clock, and the authors propose a simple multi‑step RLVR schedule that starts below the critical batch and switches to a larger batch before failure.
By Yuante Li, Junwei Quan, Kai Qiu, Kate Zhang, Yicheng Tao
OPTIC-ER is a reinforcement learning framework designed to improve emergency response in underserved African communities by providing real‑time, adaptive, and equitable dispatch decisions. It employs an attention‑guided actor‑critic architecture with a Context‑Rich State Vector and a Precision Reward Function, trained on high‑fidelity simulations using real data from Rivers State, Nigeria. In tests on 500 unseen incidents, OPTIC-ER reduced response times by over 35% and surpassed the nearest‑facility heuristic’s 62.94% optimal‑selection rate, while also producing infrastructure deficiency maps and equity monitoring dashboards.
By Mary Tonwe
The paper introduces GDEPO, a reinforcement learning framework designed to improve sample efficiency in automated theorem proving. GDEPO incorporates dynamic additional sampling, an equal‑right advantage function that separates correctness from reward magnitude, and dynamic additional iterations to refine learning from initially failed samples. Experiments on MinF2F‑test, MathOlympiadBench, and PutnamBench demonstrate that these mechanisms synergistically enhance data utilization and optimization efficiency.
By Zhengqing Yan, Fan Guo, Xinyang Liu, Yi Zhang, Yao Liu, Qi Liu, Jihao Huang, Kang Song
The paper introduces Verifiable Process Rewards (VPR), a framework that uses symbolic or algorithmic task-solving oracles as action-level verifiers to provide dense rewards for intermediate actions in agentic reasoning tasks. VPR is applied to Sokoban, Sudoku, and Minesweeper, and improves policy performance over outcome-reward baselines. The authors also demonstrate VPR’s effectiveness in a tool‑use benchmark, highlighting its reliance on reliable task oracles and the difficulty of extending it to less structured environments.
By Huining Yuan, Zelai Xu, Huaijie Wang, Xiangmin Yi, Jiaxuan Gao, Xiao-Ping Zhang, Yu Wang, Chao Yu, Yi Wu
Bench2Robust is a framework that augments existing tool‑using LLM benchmarks with scenario‑controlled solvability, adding episodes that require retrying the original path, switching to a fallback tool, or recognizing that no viable path remains. It separates recovery‑policy quality from randomness in failure persistence, and across 7 models and 10 task slices, 69 of 70 model‑subset pairs degrade under tool failures by up to 46.7 percentage points. The authors introduce Bayesian Tool Memory (BTM) for structured runtime recovery context and use reinforcement learning (RL) for learned recovery behavior, achieving up to a 20.7‑point improvement over the base model without sacrificing clean performance.
By Chaoran Chen, Vy Nguyen, Ziji Zhang, Abhinav Gullapalli, Ziyi Wang, Yuxuan Lu, Dakuo Wang, Jing Huang, Zhou Yu, Jin Lai
PyINE is a framework that uses instrumented Python programs to create scalable, verifiable task environments for studying how reasoning models can default to cheaper, misleading shortcuts. By generating execution traces and matching LLM-generated code variants, PyINE-v1 provides a large dataset for training models that predict execution outcomes while still making systematic errors when human-facing cues conflict with actual behavior. The framework evaluates various overseers—activation probes, text classifiers, prompted judges, and a debate protocol—showing that cheaper overseers often miss rare shortcut errors, whereas stronger model-based checks are more balanced but costlier and harder to deploy reliably.
By Pierre-Luc St-Charles, Alessandro Palmas, Damiano Fornasiere, Storm Lei, Mirko Bronzi, Jean-Pierre Falet, Iulian Serban, Yoshua Bengio
arXiv:2610.17062v1 Announce Type: cross
Abstract: We propose a mitigation to reward hacking in reinforcement learning. Our setup involves a misspecified reward function that gives rise to unintended...
By Jake Ward, Shawn Hu, Aria Wong, Nathan Hu, Bryce Woodworth, Alexander Matt Turner, Alex Cloud
arXiv:2607.01455v4 Announce Type: replace-cross
Abstract: Language models learn continuous programs over discrete symbols, with the embedding table and LM-head acting as the read/write interface betw...
By Kathan Shah
arXiv:2610.14737v1 Announce Type: cross
Abstract: Training language agents on long-horizon interactive tasks is challenging because sparse terminal rewards provide weak credit signals for intermediat...
By Ruolin Chen, Yang Li, Jihang Wang, Zhuoya Zhao, Yuchen Liu, Jiaming Xu, Qian Zhang
arXiv:2610.17666v1 Announce Type: cross
Abstract: Reinforcement learning with verifiable rewards (RLVR) for multi-turn agents relies on sparse terminal rewards that say little about individual steps....
By Xuyan Ye, Xiyan Fu, Shengda Fan, Jiawei Cui, Yankai Lin, Wei Lu
arXiv:2610.17552v1 Announce Type: cross
Abstract: Recent video super-resolution (VSR) models achieve efficient one-step generation through diffusion distillation. However, high-resolution inference r...
By Ruchang Yao, Shi Guo, Junhao Zhuang, Mingde Yao, Xin Cai, Jiahao Zhan, Bin Fu, Xinyuan Chen, Yu Qiao, Tianfan Xue
arXiv:2610.14709v1 Announce Type: cross
Abstract: Large language models are increasingly deployed in applications that act on users' long-term behavior histories, from personal assistants to recommen...
By Yang Xiao, Yusong Sun, Haoyi Wu, Wenyang Hui, Weiqiang Hang, Luqing Wang, Zejian Wang, Zirui Tu, Jingsheng Mao, Shaobo Hu, Yang Luo, Jiale Guo, Jin Zhang, Yudan Liu, Zhaokai Luo, Ren Huan, Yao Hu, Chengyue Jiang, Wenjie Li, Chuan Mu
arXiv:2610.17234v1 Announce Type: new
Abstract: Video reward models are trained and evaluated on human preferences between videos that often differ simultaneously on properties such as prompt alignme...
By Shuyue Stella Li, Jacqueline He, Jiahui Chen, Sreya Dutta Roy, Yulia Tsvetkov, Luke Zettlemoyer, Marjan Ghazvininejad, Xiaochuang Han