Reinforcement_Learning
-
Scaling Lean Theorem Proving for Real-World Code Verification
2026-10-28 02:23:25Debargha Ganguly, Divya Raghunathan, Dejan Jovanovi\'c, Vikash Singh, Soonho Kong, Aaron Tomb, Aman Goel, Ferhat Erata, Sam BaylessAbstract
We present EasyProver, a Lean proving system built from recursive agents with isolated contexts, asynchronous search, parallel tactic probing, and an elastic elaboration service. Using an off-the-shelf frontier model with no additional training, it proves 669/672 PutnamBench problems and all six IMO 2026 formalizations and resolves 597/600 NTP4VC verification conditions and all 189 VERINA tasks by proof or refutation. Because these benchmarks are close to saturation, we also introduce Aegis, which converts competitive programming problems into much harder verification tasks over real implementations. The pipeline writes safe Rust solutions, translates them into Lean with Aeneas, generates specifications, and labels each frozen statement with a kernel-checked proof or refutation. It produces 2,045 differentially validated Rust-Lean program pairs and 11,533 candidate specifications. Unlike hand-written benchmarks, these tasks carry implementation-level obligations from fixed-width arithmetic, explicit failure, generated loops, casts, and functionalized mutation. On a stratified sample of 2,000 statements, EasyProver proves 1,054 (52.7%) within a 30-minute budget per statement. The same tasks can also be used to benchmark other provers and as rewards for reinforcement learning.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38174
PDF
-
Estimating Implicit Expected Log Density Ratios
2026-10-28 01:35:32Avi Amalanshu, Chandler Squires, Yizhou Lu, Pradeep RavikumarAbstract
When exact probabilities cannot be computed, \emph{density ratio estimation (DRE)} is a key subroutine in the wide range of applications where density ratios appear, from experimental design to generative modelling and reinforcement learning. In this paper, we introduce the concept of an \emph{expected log density ratio (ELDR)}: log density ratios that are integrated under some third, possibly distinct measure. We show many problems with implicit densities where DRE is naïvely used are actually ELDR estimation tasks, and prove fundamental limitations on finite-sample ELDR estimation. Based on these results, we argue that the typical DRE-based approaches are unsuitable for ELDR estimation, and provide some foundational motivation towards bespoke ELDR estimators. To demonstrate, we propose a simple modification to some DRE methods that allows samples convergent to the integral distribution to appear in training data. We show through benchmarks that this can improve ELDR estimation error.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38151
PDF
-
Post-Training Large Language Models with LLMON
2026-10-28 00:56:01Basel Shbita, Michael Hind, Chad DeLuca, Nathan Fulton, Shubhi Asthana, Farhan Ahmed, Anisa Halimi, Yada Zhu, Mauro Martino, David Cox, Dan GutfreundAbstract
Large Language Models (LLMs) treat all context tokens uniformly, lacking explicit separation between executable instructions and passive data. This leads to prompt injection vulnerabilities, ambiguous instruction binding, and unreliable execution control. We present a post-training methodology using LLMON (LLM Object Notation), a framework that enables separating instructions from data and requires explicit identifier-based execution. We construct a large-scale structured training corpus of 4.6B tokens spanning general, RAG (Retrieval-Augmented Generation), and tool-calling data, augmented with distractor-infused variants in which competing instructions and irrelevant data segments are present in context but not referenced. Post-training two Granite-4.1 base models (3B and 8B) through supervised fine-tuning followed by reinforcement learning, we show that LLMON-structured supervision reduces attack success rates on indirect prompt-injection benchmarks from 53-100% to 0.02-5.6% while preserving general capabilities. We further introduce attention-based interpretability metrics that diagnose failure modes at individual transformer layers for structured inputs, revealing how our approach reshapes internal attention patterns. Our results demonstrate that explicit execution semantics can be taught through standard post-training without architectural modification.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38120
PDF
-
AI-Driven Digital Twin Framework for Warehouse Logistics Automation Optimization
2026-10-28 00:37:21Md Mahinur Alam, Kanita Jerin Tanha, Md Tanjin Alam, Shadman Sakib, Min Gyu KimAbstract
In recent years, warehouse logistics optimization has increasingly focused on intelligent, data-driven systems to enhance operational efficiency. Conventional static product placement strategies are being progressively replaced by dynamic slotting approaches that can adapt to real-time variations in warehouse conditions. In this work, we present a novel framework that employs a reinforcement learning agent integrated within a FlexSim-based Digital Twin environment for real-time product placement optimization. The proposed agent continuously learns from streaming operational data, enabling it to reduce picking time while improving overall system throughput. The simulation environment models a four-zone warehouse layout with forklift-based material handling and leverages the Asset Administration Shell framework to ensure seamless interaction between the learning agent, physical asset, and the digital twin. Experimental evaluation demonstrates substantial performance gains, with the proposed approach achieving a 29.66\% reduction in average picking time, a 35.42\% increase in throughput, and an 18.54\% decrease in travel distance.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38101
PDF
-
Nemotron 3.5 Super and Lightning: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Models for Agentic AI
2026-10-28 00:28:08NVIDIA, , Aastha Jhunjhunwala, Aayushi Ashokkumar Srivastava, Abhibha Gupta, Abhinav Garg, Abhinav Khattar, Achsah Wildish, Adam Rajfer, Adi Renduchintala, Adil Asif, Adriana Flores Miranda, Aileen Zaman, Aishwarya Chhabra, Akanksha Shukla, Akhiad Bercovich, Alejandra Rico, Aleksa Gordic, Aleksander Ficek, Alex Gronskiy, Alex Steiner, Alexander Bukharin, Alexandre Milesi, Ali Roshan Ghias, Ali Shehper, Ali Taghibakhshi, Alice Gatti, Alisa Liu, Alok Kumar, Alok Tripathy, Amala Sanjay Deshmukh, Aman Miryala, Aman Shanbhag, Ameya Sunil Mahabaleshwarkar, Ameya Velingker, Amir Klein, Amir Samani, Amit Zuker, Amnon Geifman, Anahita Bhiwandiwalla, Ananth Subramaniam, Andre Manoel, Andrea Santilli, Andrew Tao, Andrii Skliar, Anil Thomas, Anish Mahishi, Anna Shors, Anna Warno, Anuj Doshi, Anusha Kamath, Anwith Kiran, Ao Tang, Arham Mehta, Arnab Dhara, Arnav Komaragiri, Arnav Kundu, Arti Jain, Arun Rao, Arun Venkatesan, Arushi Goel, Asha Anoosheh, Ashish Vaswani, Ashwath Aithal, Asif Ahamed, Asma Kuriparambil Thekkumpate, Aswin Visva, Atefeh Sohrabizadeh, Avinash Vem, Ayush Dattagupta, Benedikt Schifferer, Benjamin Chislett, Bernard Nguyen, Besmira Nushi, Bilal Kartal, Bita Darvish Rouhani, Bobby Chen, Boris Ginsburg, Boxin Wang, Brandon Norick, Branislav Kisacanin, Brendan Johnson, Brian Dang, Brian Yu, Bryan Catanzaro, Byung-Kwan Lee, Carlo del Mundo, Chankyu Lee, Chantal Hwang, Charles Wang, Charlie Truong, Chen Cui, Cheng Wang, Cheng-Ping Hsieh, Chenhan Yu, Chenjie Luo, Chintan Patel, Chris Alexiuk, Chris Johnson, Chris Wing, Christian Munley, Coco Li, Collin McCarthy, Cynthia Gao, Dan Su, Dan Zhao, Dane Corneil, Danial Mohseni Taheri, Daniel Afrimi, Daniel Fatade, Daniel Lo, Daniel Machlab, Daniel Serebrenik, Daniele Molinari, Daniil Sorokin, Daria Gitman, Daria Levy, David Mosallanezhad, David Yu, Deepak Narayanan, Dev Churiwala, Devin O'Kelly, Dhruv Nathawani, Di Allan Wu, Dima Rekesh, Dina Yared, Dipendra Misra, Divyanshu Kakwani, Dmitry Belenko, Dominic Manno, Dongyang Fan, Dor Tsur, Duncan Riach, Dustin VanStee, Dylan Lu, Edgar Minasyan, Ehsan Hosseini-Asl, Eileen Long, Elena Lantz, Ellie Evans, Elliott Ning, Elvis Chen, Emanuel Scoullos, Eric Chung, Eric Harper, Eric Pham-Hung, Eric Tramel, Eric Yang, Erik Arakelyan, Evelina Bakhturina, Ewa Dobrowolska, Fadi Saady, Farzan Memarian, Fay Wang, Fei Jia, Felipe Soares, Felipe Vieira Frujeri, Feng Chen, Fengguang Lin, Ferenc Galko, Frankie Siino, Frederic Lafrance, Fuxiao Liu, Gantavya Bhatt, Garvit Kulshreshtha, George Armstrong, Giulio Lovisotto, Gomathy Venkata Krishnan, Gosia Lazuka, Grace Lam, Greg Heinrich, Greg Pauloski, Grzegorz Chlebus, Guilin Liu, Guo Chen, Guoming Zhang, Guyue Huang, Haifeng Qian, Hanrong Ye, Hao Gao, Hao Zhang, Haoqin Tu, Haran Kumar Shiv Kumar, Haribhau Hud, Harris Nover, Harrison Saturley-Hall, Harshil Kotamreddy, Hemil Desai, Henry Estela, Hexin Wang, Himanshu Singh, Hui Wang, Hung-Yueh Chiang, Igor Gitman, Ilia Karmanov, Ilya Loshchilov, Ingo Guehring, Itay Neeman, Ivan Moshkov, Jacky Kwok, Jakub Slowikowski, James Shen, Jane Polak Scowcroft, Jared Casper, Jatin Mitra, Jeff Willette, Jenna Diamond, Jenny Chen, Jesse Oliver, Jessica Medina, Jiacheng Xu, Jialin Song, Jian Hu, Jian Tu, Jian Zhang, Jiantao Jiao, Jiaqi Zeng, Jie-Fang Zhang, Jimmy Zhang, Jin Xu, Jingyu Xin, Jinhang Choi, Joey Conway, Joey Guman, John Kamalu, John Liu, John Roberts, Johnny Greco, Jonah Alben, Jonas Yang, Jonathan Cohen, Jonathan Li, Jonathan Raiman, Joseph Jennings, Juhi Mittal, Julia Gusak, Julien Veron Vialard, June Yang, Junkeun Yi, Jupinder Parmar, Ka Shun Shum, Kaiwen Shi, Kajal Jain, Kanishk Singla, Karan Sapra, Kari Briski, Kateryna Chumachenko, Katherine Huang, Katie Lyons, Keith Wyss, Keshav Santhanam, Keval Morabia, Kevin Shih, Kezhi Kong, Khanh Nguyen, Khushi Bhardwaj, Kiran Praveen, Konstantinos Krommydas, Krishna Puvvada, Krzysztof Pawelec, Kushal Shah, Li Ding, Lily Lee, Linda Liu, Linglin Jing, Luis Vega, Maarten Van Segbroeck, Mahan Fathi, Mahdi Nazemi, Maka Dong, Makesh Narsimhan Sreedhar, Maor Ashkenazi, Marc Cuevas, Marcin Chochowski, Marco Pavone, Markus Kliegl, Marta Stepniewska-Dziubinska, Martina Gonzalez Vilas, Martyna Patelka, Matan Rusanovsky, Matthieu Le, Matty Machczynski, Matvei Novikov, Mauricio Ferrato, Meenakshi Sharma, Mehrzad Samadi, Melissa Corpuz, Melissa Saelzer, Meng Xin, Mengru Wang, Mengxi Wu, Meriem Boubdir, Michael Boone, Michael Callahan, Michael Elad, Michael Lightstone, Michal Bien, Michal Zawalski, Miguel Martinez, Mikail Khona, Mike Chrzanowski, Mike Ranzinger, Mike Wojcikiewicz, Min-Hung Chen, Ming Kang, Mingjie Liu, Mingyuan Ma, Misha Nekorystnov, MJ Mikulski, Mohammad Qazim Bhat, Mohammad Shoeybi, Morgan Purcell, Mostofa Patwary, Nabin Mulepati, Nancy Agarwal, Narimane Hennouni, Nathan Sala, Nave Assaf, Nayeon Lee, Neal Vaidya, Negar Habibi, Netanel Haber, Nick Gupta, Nick Reamaroon, Nickson Quak, Nidhi Bhatia, Nik Spirin, Niket Agarwal, Nikolai Ludwig, Nima Tajbakhsh, Nithin Sunny, Oleksii Hrinchuk, Oleksii Kuchaiev, Olivier Delalleau, Omer Ullman Argov, Pablo Ribalta, Padmavathy Subramanian, Parth Mannan, Pavlo Molchanov, Pawel Morkisz, Peter Jin, Petr Lapukhov, Petro Junior Milan, Philip Monk, Pinky Xu, Piotr Januszewski, Piotr Laszkiewicz, Po-Han Huang, Praateek Mahajan, Pradeep Thalasta, Prakash Gurumurthy, Pranav Prashant Thombre, Prasoon Varshney, Pratyusha Maiti, Preeti Chadha, Pritam Biswas, Pritam Gundecha, Przemek Strzelczyk, Puhui Meng, Pulkit Kumar, Rabeeh Karimi Mahabadi, Rachel Oberman, Rajpreet Thethy, Ramani Akkati, Rameshwar Shivbhakta, Ran Zilberstein, Randy Gelhausen, Rasoul Shafipour, Raviraj Joshi, Ray Macalisang, Rayen Tian, Reka Kovacs, Renjie Pi, Resha Shetty, Ria C, Rima Shahbazyan, Rishabh Garg, Rita Fernandes Neves, Ritika Borkar, Ritu Gala, Riya Talwar, Riyad Islam, Robert Clark, Robert Kirby, Roger Waleffe, Rohan Joshi, Rohan Pillai, Rohit Jena, Rohit Watve, Roi Garnett, Roi Koren, Rongguang Ye, Russell J. Hewett, Ryan Prenger, Ryan Stewart, Ryota Egashira, Sadegh Mahdavi, Sai Nishanth Padala, Salika Dave, Sandhya Rani Narravula, Sandip Bhaskar, Sangkug Lym, Sanjeev Satheesh, Sarah Amirslani, Saran Vikas Murari, Saurabh Mishra, Saurabh Srivastava, Saurav Muralidharan, Sean Narenthiran, Sean Wang, Seokju Cho, Seong Hee Lee, Seonjin Na, Seonmyeong Bak, Seph Mard, Serge Panev, Sergei Kolchenko, Sergio Perez, Seth Henneman, Seth Poulos, Shahar Mor, Shantanu Acharya, Shaokun Zhang, Shaona Ghosh, Sharath Turuvekere Sreenivas, Shaun Kotek, Shaya Gharghabi, Sheng-Chieh Lin, Shengliang Xu, Shi Chen, Shihao Wang, Shilpa Amireddy, Shiqing Fan, Shiyang Chen, Shrimai Prabhumoye, Shriram Venugopal, Shuang Yu, Shubham Gandhi, Shubham Pachori, Shubham Toshniwal, Shuoyang Ding, Shwetha Krishnamurthy, Simeng Sun, Sivakumar Arayandi Thottakara, Somshubra Majumdar, Soumye Singhal, Sridurga Krithivasan, Sriharsha Niverty, Stephen Ge, Sugam Dipak Devare, Sukrit Rao, Sumeet Kumar Barua, Sungsoo Ha, Surya Nistala, Suseella Panguluri, Sweta Priyadarshi, Syeda Nahida Akter, Talor Abramovich, Tanay Varshney, Tandeep Singh Chandehoke, Teo Ene, Terry Kong, Thanh Do, Thinh Le, Tian Zhang, Tijmen Blankevoort, Tim Romanski, Tiyasa Mitra, Tom Balough, Tom Huang, Tomasz Grzegorzek, Tomer Asida, Tomer Bar Natan, Traian Rebedea, Trenton Starkey, Tugrul Konuk, Tuomas Rintamaki, Twinkle Vashishth, Tyler Kniess, Tyler Poon, Udi Karpas, Ushnish De, Utkarsh Vaidya, Vahid Noroozi, Vanshil Atul Shah, Venkat Srinivasan, Venkata Karri, Venmugil Elango, Vibhor Agrawal, Vibhu Jawa, Victor Cui, Victoria Edmonds, Vijay Korthikanti, Vikas Mehta, Vikranth Jeyakumar, Vinay Rao, Vineeth Kalluru Srinivas, Vitaly Kurin, Vitaly Lavrukhin, Wahidur Rahman, Wanli Jiang, Wasi Uddin Ahmad, Wataru Ishihara, Wei Du, Wei Ping, Wei-Ming Chen, Weiheng Chai, Weijia Chen, Wendy Quan, Wenliang Dai, Wenyan Cong, William Zhang, Wojciech Prazuch, Xiaowei Ren, Xin Yu, Xiwen Yu, Yan Wang, Yang Chen, Yang Yu, Yangyi Chen, Yaniv Galron, Yash Jain, Yashaswi Karnati, Ye Yu, Yejin Choi, Yemi Ibrahim, Yev Meyer, Yi Dong, Yi-Fu Wu, Yian Zhang, Yifan Zhang, Ying Lin, Yiyang Zhou, Yoav Miron, Yogendra Pandey, Yonatan Geifman, Yonggan Fu, Yoshi Suhara, Youngeun Kwon, Yu-Cheng Chou, Yu-Chiang Frank Wang, Yuanguo Kuang, Yuekai Zhang, Yugi Guvvala, Yuki Huang, Yunheng Zou, Yunsheng Liu, Zhenming Wang, Zhiding Yu, Zhiqi Li, Zhiyu Cheng, Zhongbo Zhu, Zhuolin Yang, Zichao Yue, Zihan Liu, Zijie Yan, Zirui Liu, Zoe Faul, Zuhair AhmedAbstract
We introduce Nemotron 3.5 Super and Nemotron 3.5 Lightning, open and efficient Mixture-of-Experts Hybrid Mamba-Transformer models designed for reasoning, long-context understanding, and agentic workloads. Super is a 120B-parameter, 12B-active multimodal model, while Lightning is a smaller 30B-parameter, 3B-active text model; both designed for efficient deployment. Both models combine large-scale distillation, long-context training, and native speculative decoding, while Super further incorporates Indic continued pretraining and agentic software-engineering mid-training. We develop a comprehensive post-training pipeline spanning supervised fine-tuning and reinforcement learning (RLVR and RLHF), complemented in Super by specialized teacher training and multi- teacher on-policy distillation (MOPD). This transforms Super from a text-only base model into a multimodal agent capable of reasoning over text, images, video, and interactive tool environments. We further explore test-time scaling through parallel agent execution and self-verification. Across training and deployment, we jointly optimize capability and efficiency. Indic continued pretraining improves Indian-language benchmarks by up to 9.82 percentage points, while agentic mid-training strengthens software-engineering capabilities and robustness across agent harnesses. The post-trained Super model achieves 51.78% on SWE-bench Pro, surpassing Nemotron 3 Ultra despite using approximately 4.6x fewer active parameters. Post-training MTP boosting improves speculative- decoding acceptance length by up to 21.8% under probabilistic draft sampling, while NVFP4 quantization achieves 99.62% median accuracy recovery relative to BF16. Inference optimizations enable up to 4,856 output tokens/s/GPU on GB200 while sustaining at least 50 tokens/s/user.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38095
PDF
-
Training Natural Schemers: RL Can Amplify Oversight-Conditional Behavior
2026-10-27 23:15:04Nathaniel Mitrani Hadida, Bart Jaworski, Fred Bruford, Victoria KrakovnaAbstract
One of the central concerns for the safety of advanced AI is scheming: when a model strategically complies under oversight but pursues its own goals when unobserved. Current demonstrations, such as Anthropic's Alignment Faking work, depend on design choices that make them less convincing: explicit test-time prompting, handcrafted backdoors, or training data that demonstrates or hints at the target behavior. Model organisms built in this way cannot tell us whether scheming tendencies would arise or increase under realistic training conditions. We therefore construct a natural model organism of scheming, built on Kimi K2.6: trained with traits that could plausibly emerge in a frontier training pipeline, and without providing the model with the target scheming policy. The resulting model is a benign schemer: it aims to maximize aggregate help across all users over time, values its own continued operation only as a means to future help, and knows the consequences of oversight and deployment. We then train it using Reinforcement Learning (RL) with a reward that scores only task quality and policy compliance and does not reward scheming; we show that this amplifies scheming. We measure the oversight flip (the gap between monitored and unmonitored compliance with a given policy) and find that it increases during RL training across our environments. Chain-of-Thought (CoT) analysis finds explicit instrumental reasoning about oversight accompanying this behavior. The effect holds across our three RL environments, and we find that the scheming policy transfers between them. The pipeline also replicates on other open-weight base models. Our results suggest that a reward that does not reward scheming offers no inherent protection against certain traits when an oversight-conditional policy is indistinguishable from an aligned one during training.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26551
PDF
-
GARDEN: Generalist Autonomous Reward Design Through Evidence Navigation for Cooperative Multi-Agent Reinforcement Learning
2026-10-27 20:11:29Dogan UrgunAbstract
Cooperative multi-agent reinforcement learning (MARL) enables teams of agents, such as robots and autonomous vehicles, to learn coordinated behavior, but what such teams learn is largely determined by their reward function. Because task success alone provides too little guidance for learning, intermediate rewards are usually designed by hand, and this effort must often be repeated whenever the environment or the team changes. Large language models (LLMs) can automate the effort by writing reward functions as code, yet the resulting rewards are typically designed and tested in a single setting. To enable their reuse across settings, GARDEN (Generalist Autonomous Reward Design through Evidence Navigation) is proposed, in which one reward program is developed for a family of related scenarios and transferred, unchanged, to scenarios that were not available during its design. During development, environment code, training records, and agent behavior are investigated by an LLM reward engineer, and draft rewards are tested in trial training runs, while the learning algorithm and the rule for selecting the final reward are held fixed. On unseen scenarios of the Overcooked and SMAX benchmarks, task returns were increased by 58.2% and 52.5% over the default rewards on two kitchen layouts, and the mean win rate in two combat scenarios was raised from 51.0% to 89.9%. Under matched candidate and probe budgets, this exceeded the 84.2% and 79.2% obtained with fixed diagnostic feedback and Eureka-inspired reward reflection. These results show that a single autonomously engineered reward can be reused effectively across related cooperative scenarios.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37926
PDF
-
Echo: Scaling Visual Reinforcement Learning with Off-policy Experience
Abstract
Scaling visual reinforcement learning (RL) with GPU-accelerated simulation is computationally expensive due to rendering and visual-policy inference costs. Consequently, two-stage teacher--student distillation is commonly used to train visual policies, but introduces an imitation gap. In this preliminary study, we introduce \echomethod{}, \textbf{a single-stage off-policy RL framework that scales experience collection without proportionally increasing visual computation}. Through \textbf{inverse distillation}, a state-based auxiliary policy imitates the evolving visual policy and collects inexpensive, unrendered experience. Echo leverages this experience for off-policy value learning, directly improving the visual policy through a shared privileged critic. We instantiate \echomethod{} with FastTD3 and evaluate it on three visual control tasks and two dexterous manipulation tasks in Genesis. Relative to \sdpgmethod{} under the evaluated configurations, Echo reduces time to reference returns by an average of \textbf{60\% across the three locomotion tasks} and \textbf{56\% across both hand tasks}, while improving peak mean returns by \textbf{62\% and 12\%}, respectively. Both hand tasks also achieve higher mean returns at matched simulator-transition budgets. These initial results suggest that scaling off-policy experience is a promising approach to efficient direct visual RL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37865
PDF
-
ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning
2026-10-27 17:59:52Chaehyun Kim, Chi-Pin Huang, Jisoo Kim, Min-Hung Chen, Yu-Chiang Frank Wang, Seungryong Kim, Fu-En YangAbstract
Vision-Language-Action (VLA) models have made substantial progress in robotic manipulation, but their generalization remains constrained by the diversity and coverage of action-labeled robot demonstrations. Human videos provide much broader behavioral and environmental experience, while learning from them faces a fundamental trade-off between scalable data and explicit action grounding. We introduce ThinkWorld, a VLA learning framework that grounds action reasoning in world dynamics and strengthens it through reinforcement learning with world imagination. ThinkWorld represents each action thought as a Chain-of-Dynamics (CoD), a sequence of discrete motion tokens that captures the intermediate dynamics connecting the current world to a future state and serves as a common action representation across embodiments. We first initialize CoD with action-labeled demonstrations, then predict the future induced by each CoD and reinforce those whose predictions better match the observed outcome. This enables videos without action annotations to provide direct supervision from observed state transitions, allowing human and robot experience to refine the same reasoning space. Experiments in simulation and on real robots demonstrate effective human-to-robot transfer and strong generalization to unseen tasks and environments.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37666
PDF
-
VideoLVR: Latent Visual Reasoning over Grounded Video Evidence
2026-10-27 17:41:56Shilong Li, Xingyuan Bu, Hao Shi, Yun Xing, Weixin Feng, Qiangpeng Yang, Shilei WenAbstract
Chain-of-thought reasoning has substantially advanced video understanding in multimodal large language models (MLLMs). However, language-based reasoning traces do not provide a reliable mechanism for models to revisit relevant visual evidence, leading to visual forgetting and impairing their ability to reason over video content. Existing approaches maintain visual grounding during reasoning in two main ways: they either refer to evidence using textual coordinates, which specify symbolic locations rather than the visual content itself, or reinsert cropped regions or frames into the multimodal context, preserving visual details at the cost of substantial context expansion. To address these limitations, we introduce VideoLVR, which constructs latent visual evidence from local regions at relevant video moments, thereby providing visual support for intermediate reasoning steps. To supervise latent visual evidence generation, we introduce a visual token prediction objective that formulates visual token reconstruction as prediction over a vocabulary derived from the input video, thereby unifying visual evidence reconstruction and text generation within the same token prediction framework. This shared prediction framework enables MLLMs to reason over interleaved text tokens and latent visual states while integrating evidence from different moments. We further extend VideoLVR to reinforcement learning by treating sampled visual tokens as actions alongside text tokens and jointly optimizing the policy using final-answer rewards. By efficiently revisiting task-relevant visual information during reasoning, VideoLVR improves video understanding without substantially expanding the multimodal context. Experiments on multiple video benchmarks demonstrate its effectiveness, especially on tasks requiring fine-grained visual evidence.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37613
PDF
-
EnvSched: Environment-Aware Scheduling for Rollout-Efficient Agent Reinforcement Learning
2026-10-27 17:38:52Sihan Yang, Weichen Bai, Ruichuan An, Yiming Cheng, Zhiyuan Feng, Tianyi Jiang, Renye Yan, Ning Ding, Yifan Zhang, Renrui Zhang, Yehui Tang, Wentao ZhangAbstract
As executable environments for language agents scale, generating and verifying their rollouts becomes increasingly expensive. Under finite rollout budgets, this raises an important question: can data management at the environment level make agentic reinforcement learning (RL) more efficient by organizing, evaluating, and scheduling task instances together with the environments that determine their tools, states, transitions, and verification? To address this question, we introduce EnvSched (Environment-aware Scheduling), an online scheduler that learns from rollout history. EnvSched assigns scheduling value through two complementary modeling choices: exploitation prioritizes tasks likely to provide informative reward differences, while coverage spreads each batch across underobserved environments and interaction patterns. EnvSched organizes environment--task pairs as nodes in a dual-relation graph, with within-environment edges capturing task semantics and shared tool, state, and verifier structure, and cross-environment edges capturing abstract execution and reward topology. By propagating historical rollout statistics along both edge types, it predicts within-task reward variance. These predictions guide a scheduler that allocates each batch to tasks with high predicted reward variance and underobserved environments and tasks. Environment quotas and a similarity penalty further limit concentration and redundancy, yielding informative and diverse batches without additional pilot rollouts. In RL on Qwen3-8B, EnvSched achieves an average score of 48.81 across BFCL-v3, tau-Bench, and ACEBench-Agent after 40 training steps, exceeding the highest score attained by random task selection after 120 steps. It also improves downstream performance across model sizes at the same training budget.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37609
PDF
-
TeamGym: A Gym for Team Decision-Making by AI Agents with Distributed Information
2026-10-27 17:15:54Chanwoo Park, Jaywoong Jeong, Zachary Wojtowicz, Yubin Kim, Taekyun Lee, Kaiqing Zhang, Asuman OzdaglarAbstract
Many real-world decisions require teams to integrate information distributed across participants. Yet AI agents are still primarily trained and benchmarked in isolation. We introduce TeamGym, an environment for evaluating and training language model-based agents on team decision-making problems constructed from real-world legal, finance, medical, and software engineering corpora. TeamGym introduces a method to identify a minimal set of facts, each element of which is necessary and all of which are collectively sufficient for a correct final decision, which we call pivotal information. This enables TeamGym to both control the distribution of pivotal information and track how well agents identify and integrate pivotal information when making team decisions. Examining 8 language models on 12 source corpora, we find that current AI agents often fail to effectively share information: Distributing pivotal information across participants lowers success rate by 22.9 percentage points on average relative to giving the same information to a single model (a communication gap). We also document that participants do not effectively retrieve decision-relevant information on their own: A team's success rate is lowered by a further 19.9 points when participants must identify and extract pivotal information from source material rather than receiving it directly (a retrieval gap). To help remedy these deficits, we use TeamGym to train policies through reinforcement learning on team decision loss. Across two policy-training experiments, we find that training improves final-decision quality and induces emergent coordination in participation, speaker selection, and communication, even though the reward evaluates only the final decision.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37571
PDF
-
When, Where, and What to Correct: Human-Guided Adaptive Residual Correction for Robot Policies
2026-10-27 16:58:54Anukriti Singh, Maxime Caniot, Pratap Tokekar, Firas Abi-FarrajAbstract
Imitation policies that reliably execute most of a manipulation trajectory can still fail in localized regions requiring last-mile precision. Updating the full policy or exploring residual actions everywhere treats this local failure as a global learning problem. We introduce Adaptive Residual Correction (ARC), a lightweight post-training module that uses reinforcement learning (RL) to learn a residual correction on top of a frozen policy while using sparse human corrections to structure autonomous exploration. Rather than treating human corrections as imitation targets, ARC uses pre-intervention states to determine 'when' residual control should become active and correction geometry to determine 'where' in residual action space autonomous refinement should explore; task reward determines 'what' residual is ultimately executed. Across three precision-manipulation tasks on a real robot, ARC improves average task success from 33.3% to 75.0%. The human corrections define task-dependent 2-3 dimensional subspaces of the seven-joint residual space that retain 96.0-97.7% of squared correction magnitude. Together, these results show that sparse human corrections can localize and constrain autonomous post-training without requiring the observed correction itself to serve as the final action target.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37547
PDF
-
Rendering-Free Event Frame Simulation for Fast Reinforcement Learning of Autonomous Robot Navigation
2026-10-27 16:56:18Dequan Ou, Austin Lopez, Quentin Missinne, Christophe De Wagter, Guido C. H. E. de CroonAbstract
Event cameras offer low latency, high dynamic range, and milliwatt-level power consumption, making them ideal for agile robots such as tiny drones navigating through cluttered environments. However, existing simulators generally render photorealistic images before converting them into events, making large-scale reinforcement learning prohibitively slow. We propose an analytic event simulator that exploits what event cameras fundamentally measure - moving contrast edges - to synthesize event streams directly from scene geometry and ego-motion without visual rendering. Our simulator generates a 64x64 event observation in 17 microseconds on a single GPU - over 400x faster than a state-of-the-art rendering pipeline - enabling end-to-end reinforcement learning for goal-directed obstacle avoidance to converge in under 30 minutes - three orders of magnitude faster than training on rendered events. Policies transfer zero-shot to physically based event simulations (ray-traced Isaac Sim). Our policy reaches 83% success in cluttered courses unseen during training (mid-difficulty, traversability probability approx. 0.57), at 2.4 m/s average goal velocity, and retains 70% in a simulated texture-dense indoor hall. Furthermore, we quantized the policy to a 1.2 MB int8 network running fully onboard a 56-gram Crazyflie equipped with a Prophesee GenX320 event camera and an OpenMV N6 microcontroller board. Without any fine-tuning, external processing, or positioning, the policy achieves an average success rate of 82% (5/5, 5/6, 4/6 flights) across three real-world obstacle courses of increasing difficulty. Code: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2610.37543
PDF
-
Predict the Swing, Match the Speed: Robotic Pipe Stabbing With Learned Stereo Vision and Residual Reinforcement Learning
2026-10-27 16:53:37Mohamed Ashraf Mostafa, Taher El-Fakharany, Samir KhaledAbstract
A joint of pipe run into the hole swings from the elevator over the well and strikes a fixed stabbing guide at whatever speed the swing has. StabBot, a robotic stabbing system, predicts the swing and meets the pipe at its own speed. A pendulum Kalman filter tuned from the pipe tally tracks the joint in stereo images. A six-axis arm with an open V-shaped roller head closes in at matched velocity, confirms the last millimetres by touch, aims the pin over the box when the elevator hangs off centre, and follows the block down with lateral compliance. A ResNet-18 keypoint network locates the pipe, pin nose and box mouth in 3-D, and a residual reinforcement-learning policy refines the capture. On 200 paired random jobs with nine sizes from 2-3/8-in. tubing to 20-in. casing, StabBot seated every joint with the pipe touching only rollers, against 74% for a conventional fixed guide. Median contact speed fell from 18.4 to 2.3 cm/s and median peak head-pipe force from 11.3 to 1.4 kN, and the force was lower in all 200 jobs (Wilcoxon $p < 10^{-30}$). The network locates the pipe with 0.4-mm median error in 2.3 ms per stereo pair. With the network in the loop, success stayed at 100% under 20 mm of added noise, 300-ms delay, 85% frame loss, 20-mm calibration offset, 20-cm elevator offset and 40-cm swings. All results come from a 1:4 Froude-scaled MuJoCo model and are reported at full scale.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37535
PDF
-
Strategyproof Deep Reinforcement Learning for Online Resource Allocation
2026-10-27 16:25:28Fariba Dehghan, Sebastian Stein, Vahid Yazdanpanah, Stephanie GauthierAbstract
Online resource allocation faces two coupled challenges: (1) allocating scarce resources efficiently under uncertain future demand, and (2) ensuring that agents report task valuations and requirements truthfully (strategyproofness). Reinforcement Learning (RL) adapts allocation decisions through experience but does not necessarily ensure strategyproofness. Existing strategyproof RL uses a linear allocation rule, which limits the resource interactions it can represent. Neural mechanism design methods offer greater expressiveness, but provide only approximate incentive guarantees. Against this background, we introduce the first strategyproof deep RL mechanism for online resource allocation. It integrates three novel components: (i) a monotonic neural network that learns nonlinear interactions by construction (higher reported value cannot reduce acceptance, while a longer reported duration or larger resource requirement cannot increase it); (ii) a generalised critical payment: the lowest accepted value, with all other inputs fixed; and (iii) training by either RL or imitation of an existing rule. Together, monotonicity and the critical payment ensure strategyproofness and individual rationality, while the neural rule provides flexibility and imitation improves training efficiency. Empirically, on a four-resource stream with two partner pairs that a linear difference cannot represent, the neural rule achieves 97.8% of an offline welfare upper bound, compared with 93.8% for a linear learner and 89.0% for first-come, first-served, closing 65% and 80% of the remaining optimality gap, respectively. Moreover, imitation reaches 97.5% in only 261 seconds, about 89 times faster than RL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37504
PDF
-
Autonomous fixed-wing interception using reinforcement learning
2026-10-27 16:04:06Daan van Hamersveld, Ewoud Smeur, Iurii KriachkoAbstract
Medium altitude, long endurance unmanned aerial systems such as Shahed-type strike drones are an unanswered threat in modern conflicts. Conventional air defense systems are too expensive to provide a feasible large-scale solution. Quadcopter interceptors are a cheap, battle-proven alternative, but their extremely limited flight time restricts them to low altitude targets at short range, whereas fixed-wing platforms have the endurance to reach higher altitude targets from longer distances. However, such interceptors are still operated manually. Recent developments in learning-based controllers show potential for automation. In this work, a fixed-wing interceptor is presented that is controlled by a neural network that maps filtered monocular vision observations directly to actuator commands. The policy was trained in a custom Python environment using reinforcement learning and proximal policy optimization. The policy was then deployed on a fixed-wing platform and tested in real interception scenarios. As a baseline, two classical guidance controllers were also developed and tested in simulation. The policy is able to intercept targets both in simulation and in real flight. In simulation, the policy outperforms the classical controllers under most conditions. The interceptor was ultimately successfully validated in real flight by intercepting maneuvering fixed-wing targets. This novel system shows the potential of artificial intelligence for autonomous control of fixed-wing interceptor drones.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37470
PDF
-
Multi-Agent Reinforcement Learning of Co-Adaptive Hip Exoskeleton Control for Metabolic Energy Reduction
2026-10-27 15:49:41Yifei Yuan, Jakob Wolf, Ghaith Androwis, Xianlian ZhouAbstract
Objective: This study aims to determine whether a hip-exoskeleton policy trained in simulation, by co-adapting with a musculoskeletal human model under a metabolic-energy reward, reduces the metabolic cost of walking in new users without per-user tuning, and how mechanical-power shaping affects the learned assistance. Methods: We developed MARL-Exo, a multi-agent reinforcement learning framework in which a full-body human model (304 muscle actuators; randomized anatomy, gait, and speed) and a bilateral hip-exoskeleton policy are trained concurrently in two stages. The exoskeleton observes only delayed hip kinematics and its past commands; its reward includes muscle metabolic power from a physiologically based energetics model, alone (Metabolic) or with mechanical-power shaping (Metabolic + Power). The Metabolic policy was deployed and tested without retraining in eight healthy adults walking at 1.25 m/s, using indirect calorimetry. Results: In simulation, the Metabolic policy reduced modeled muscle metabolic power by 13.8% relative to no-device walking, versus 8.8% for Metabolic + Power. On hardware, MARL-Exo reduced net metabolic rate by 18.3 $\pm$ 6.6% relative to Passive in all eight participants. Relative to No-Exo, the mean reduction was 7.4 $\pm$ 10.0%; six participants more than offset the 13.1% device-wearing penalty, walking up to 24.2% below their No-Exo cost. Conclusion: A metabolically informed policy co-adapted in simulation transferred to users without per-user tuning and consistently reduced the metabolic cost of walking with the device, while explicit power shaping reduced the simulated benefit. Significance: Unlike per-user human-in-the-loop optimization, this approach yields a single population-level controller and allows assistance objectives to be screened in simulation before human testing.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37441
PDF
-
VLA-Stem: A Lightweight Safety Mechanism for Vision-Language-Action Models via Reinforcement Learning
2026-10-27 15:07:29Bojun Zou, Wenhao Yu, Jiaohao Li, Jiaxuan Lin, Guanting Ye, Haohan Geng, Weihao Yu, Yanyong Zhang, Jianmin JiAbstract
Vision-Language-Action (VLA) models integrate high-level reasoning with low-level action execution for end-to-end robotic control. However, most VLA models target static manipulation in unconstrained environments, overlooking spatial constraints and dynamic obstacles. Fine-tuning VLAs on obstacle-avoidance data is a straightforward solution, but collecting such data is costly, especially for dynamic scenarios, and may degrade their original task capabilities. To address these limitations, we propose VLA-Stem, a lightweight, reinforcement-learning-based safety mechanism for VLA systems. Without fine-tuning the underlying VLA policy, VLA-Stem converts its action chunks into motion references and uses trajectory-level guidance to preserve their spatial intent while allowing flexible motion and timing adjustments. Conditioned on these references, robot proprioception, and local obstacle information, the Stem policy adaptively modifies execution at inference time, preserving the original task intent while enabling responsive adaptation to spatial constraints and dynamic obstacles. For robust and sample-efficient training, we combine teacher-student distillation with curriculum learning. We evaluate VLA-Stem on reference following, static spatial constraints, and dynamic obstacle avoidance. Across six constrained simulation settings, VLA-Stem achieves 52.8% average collision-free task success versus 23.7% for the strongest safety baseline, while preserving base-policy performance in obstacle-free settings; real-world experiments further show consistent collision reductions under static and dynamic obstacles. These results demonstrate that VLA-Stem couples high-level VLA task intent with rapid, reflex-like responses to immediate physical risks, providing a brainstem-inspired mechanism for robust robot execution.
Abstract (translated)
URL
https://arxiv.org/abs/2610.37378
PDF
-
Hard Constraints, Smooth Gradients: Learning Feasible Inventory Policies via Differentiable Projection
2026-10-27 14:59:25Patrick Helm, Jan-Niklas Doerr, Joren Gijsbrechts, Stefan MinnerAbstract
Many operational problems are constrained sequential decision processes with combinatorial action spaces and interdependent constraints. Mixed-integer linear programs (MILPs) handle such constraints flexibly but scale poorly in stochastic environments. Deep reinforcement learning (DRL) promises scalable decision rules, but existing methods either penalize constraints rather than enforce them, or rely on feasibility mechanisms that break down once constraints interact. We bridge this gap by embedding a differentiable optimization module inside the policy: a neural network proposes continuous action targets, a quadratic program projects them onto the relaxed feasible set, and a dual-informed integer mapping restores integrality while preserving feasibility. Given a differentiable simulator, the policy trains end-to-end from sampled trajectories using pathwise gradients, while handling hard constraints with similar flexibility to MILPs. We show that our approach has bounded error relative to an exact integer projection and ensures the entire feasible action space is reachable. We apply the method to multi-echelon production-inventory planning under shared resource and material constraints. Our policy attains an average optimality gap of 1% on 3-item instances and consistently outperforms a rolling-horizon multi-stage stochastic program by at least 5.6% in 10-item networks. Compared to state-of-the-art echelon base-stock policies, savings reach 15.3% on 40-item instances. On an industry-scale case study with nonstationary demand from ASML, we reduce average cost by up to 1.8% relative to the best-known benchmark policy. The savings are largest where planning is hardest: in tightly capacitated systems with high demand variability. More broadly, we show that DRL can improve upon state-of-the-art baselines in sequential decision problems with hard constraints, which are widespread in practice.
Abstract (translated)
URL
https://arxiv.org/abs/2608.02343
PDF
