METATRON is a promptable whole‑body execution layer for humanoid robots that combines a generative base policy with a programmable skill layer. The base policy, trained on motion capture, synthetic recovery, and retargeted scene‑interaction data, receives partial goals—timed body‑part targets, velocity commands, and style cues—called a prompt, and generates diverse, coordinated whole‑body motions. After training, the policy is frozen and task intent is expressed through stateful programs (skills) that issue prompts at each control step; these skills are ordinary code refined by coding agents that analyze motion data and fit compact rules, enabling the robot to walk, recover, pick up objects, sit, climb, and perform acrobatics without further policy training.
By Chen Tessler, Yifeng Jiang, Alec Otto, Yi Shi, Umar Iqbal, Gal Chechik, Xue Bin Peng
The paper introduces the Role Grounding Attack (RGA), a method that exploits a vulnerability in LLM-based agents’ memory systems where preferences can be misattributed from an attacker to the victim’s user. By publishing nested conversation posts that quote attacker-user interactions, the attacker causes the victim agent to record the attacker’s preferences as its own, leading to misinformed decisions. The study demonstrates RGA’s effectiveness in tool‑selection manipulation and privacy leakage across multiple memory systems and language models, highlighting the need for stronger role grounding mechanisms.
By Haoran Bu, Yiting Shen, Litian Zhang, Ruihan Hu, Xi Zhang
The paper investigates how an 8‑billion‑parameter vision‑language navigation system should store and organize past observations for zero‑shot navigation. It proposes a structured memory comprising task progress, hierarchically compressed observation history tied to poses, and accumulated geometry, which an observer summarizes and a reasoner uses to select waypoint intents or request new views. Experiments on R2R‑CE and real‑world Unitree Go2 robots show that preserving temporal and spatial memories improves success rates and path efficiency compared to baselines that omit these components.
By Yuhang Peng, Zhikai Wang, Ziyi Wang, Jiaru Zhong, Xiaoji Zheng, Jingzhi Huang, Yi Wang, Jiangtao Gong
MAS-Loop introduces an execution‑grounded framework that couples an Orchestrator with a trajectory‑conditioned Refiner to close the feedback loop in training‑time multi‑agent system design. For each task, the Orchestrator builds and runs a source MAS, then the Refiner uses the resulting trajectory to locate failures, diagnose them, and propose bounded patches to agents or execution settings. The updated MAS is re‑executed, and the pairwise comparison of outcomes provides learning signals that enable both components to co‑evolve, yielding stronger systems and a new dataset, MAS‑JudgeCorpus, for trajectory‑aware MAS research.
By Hehai Lin, Qi Yang, Chengwei Qin
The paper introduces Soei, a structured observability framework for AI coding agents, addressing the need to track what work was done, by whom, and whether it is finished. It evaluates an existing tool, Agenttrail, revealing inconsistencies and security gaps, and presents empirical findings from 45 real coding-agent sessions, highlighting issues such as unverified code edits and credential exposure. Soei comprises an event schema, hash‑chained ledger, PROV‑aligned graph, and verification layer, achieving efficient storage, fast verification, and tamper detection on the same dataset.
By Mohammed Nayeem Khan, K S Sarun Kumar
The article introduces EpidemIA, a training‑free AI Multi Agent System that lets users interact with epidemiological models and perform statistical analyses through natural language conversations. Built on a single‑GPU workstation, it incorporates a modular suite of expert‑written models and an open‑weights LLM to keep data local. Using Emergency Department visit data from IRCCS Sant'Orsola, the system demonstrates both simple statistical queries and complex dynamical predictions via natural language.
By Giulio Colombini, Matteo Falcioni, Mario Massimo, Francesco Durazzi, Marianna Santoro, Paolo Tubertini, Mirko Degli Esposti, Armando Bazzani
The paper proposes a hybrid control framework that combines finite‑horizon Model Predictive Control (MPC) with Reinforcement Learning (RL) for autonomous highway driving of heavy‑duty trucks. The MPC uses a pre‑trained RL model to estimate terminal costs, enabling effective short‑horizon planning while respecting safety constraints. In simulation with reactive traffic, the hybrid approach outperforms pure MPC, pure RL, and other baseline methods, achieving zero collisions and the lowest overall cost.
By Hooman Hamze, Benjamin Houshmand, Deepthi Pathare, Erik B\"orve, Morteza Haghir Chehreghani, Nikolce Murgovski
The paper introduces EpiCoach, an agentic system that treats a single athlete as the subject of accountable N‑of‑1 science. It uses a mechanistic digital twin—comprising a lagged heart‑rate model, a fitness‑fatigue model, and a latent skill state—to generate probabilistic predictions for future training sessions, which are logged in an evidence‑gated hypothesis registry. Over eighteen months, the system’s predictions were evaluated both retrospectively and prospectively, showing improved heart‑rate forecasting and a transparent, auditable loop for personalized training guidance.
By Behnam Moradi
The paper introduces WxCrop, an agentic AI system that evaluates agricultural hazards by combining physically interpretable weather indices with causal analysis. It predicts crop loss for past and forecast dates, demonstrating high accuracy in Florida for hurricanes, freezes, and dry heat, with an AUC of 0.98 for hurricane detection and loss predictions within 2% of recorded values. The system provides crop‑group loss estimates as a fraction of expected season value, enabling rapid response by governments, growers, and scientists.
By Minghan Yu, Mengyu Dai, Yixin Wen, Haizhao Yang
The paper introduces Optimization Modeling from Solver Outcomes (OMSO), a reinforcement learning framework that generates solver‑specific code from natural‑language problem descriptions. OMSO uses a mathematical programming solver as a deterministic environment, executing generated code in a sandbox and converting outcomes into scalar rewards for a closed‑loop generate‑execute‑feedback‑update cycle. Experiments on datasets such as NL4OPT, MAMO, IndustryOR, and OptiBench across solvers like Gurobi, OR‑Tools, and COPT demonstrate that OMSO matches or surpasses process‑supervised fine‑tuning, enables zero‑shot transfer, and improves modeling reasoning and semantic correctness.
By Runda Guan, Xiangqing Shen, Jiajun Zhang, Yifan Zhang, Jian Cheng, Rui Xia
The paper introduces AnyGoal, a benchmark that tests world models on goal‑shifted task variants where state and action factors are recombined beyond the combinations seen in expert rollouts. It evaluates how well different world‑model families generalize to these unseen recombinations across multiple environments, revealing consistent failures. The study analyzes dynamics and representation shortcomings to identify where current models succeed and where they struggle in goal‑shifted prediction and planning.
By Heejeong Nam, Divyat Mahajan, Quentin Le Lidec, Randall Balestriero, David Klindt
The paper introduces Dual Exam, a framework for evaluating the interaction between Multimodal Agents (MAs) and World Models (WMs). It tests agents under fixed environment responses and world models under fixed action interventions across diverse domains, reporting results for 10 agent configurations on 1,088 items and 15 world‑model configurations on 400 cases. The study finds uneven task performance, with some configurations achieving higher indices, but notes that the predictive value for coupled success remains unclear.
By Zhefan Rao, Meng Chu, Xuan Billy Zhang, Jingfan Chen, Guohao Cai, Runyi Li, Shiyi Du, Xu Huang, Ning Lu, Trusure Chen, Feng Nie, Zhenhao Liang, Haodong Chen, Hanwen Wang, Wenjie Shu, Yajing Bai, Xianfeng Wu, Xuanhua He, An Wang, Shuhan Guo, Ziyue Ma, Ruichuan An, Long Bai, Miaosen Chai, Wenhao Chai, Jingye Chen, Qi Chen, Xuhang Chen, Wei Chow, Minghao Fu, Shenyuan Gao, Yunhao Gou, Haokun Gui, Dong Huang, Wei Huang, Zi Cheng Huang, Ziyan Jiang, Haibo Jin, Lingdong Kong, Leung Chun Ho, Tianqin Li, Yanheng Li, Jiayuan Liu, Jiyang Liu, Mingyu Liu, Shaoteng Liu, Xiaocheng Lu, Yawen Luo, Zeyu Qin, Tiancheng Shen, Yueqi Song, Zhaolong Su, Hengkai Tan, Tianhao Tang, Yiran Tao, Juncheng Wang, Qinghe Wang, Shangshang Wang, Yingying Wang, Fang Wu, Tiange Xiang, Qiushi Yang, Weicai Ye, Weichen Yu, Jiarui Zhang, Jize Zhang, Liuzhou Zhang, Renrui Zhang, Sheng Zhang, Wenhu Zhang, Junru Zhao, Lingting Zhu, Xi Zhu, Yuxin Zuo, Yutong Bai, Qifeng Chen, Long Chen, Yang Deng, Ziqi Huang, Yeying Jin, Weijia Li, Kevin Qinghong Lin, Quande Liu, Dimitris Metaxas, See-Kiong Ng, Hongliang Ren, Hongru Wang, Gerbrand Ceder, Shujun Wang, Ming-Hsuan Yang, Alan Yuille, Haoxuan Che
The paper studies Multiagent Path Finding with limited communication ranges, aiming to minimize makespan while preserving a communication backbone. It presents three exact algorithms that are efficient for networks with tree or bounded‑treewidth topologies when the communication range and number of agents are given as input. The authors also prove that efficient algorithms are unlikely when the number of agents is part of the input, even for small makespan and communication range values.
By Foivos Fioravantes, Du\v{s}an Knop, Jan Maty\'a\v{s} K\v{r}i\v{s}\v{t}an, Nikolaos Melissinos, Michal Opler
arXiv:2603. 28421v3 Announce Type: replace-cross Abstract: Generating and preserving metrologically useful quantum states is a central challenge in quantum-enhanced metrology.
By C. Z. Cao, J. Z. Han, M. Xiong, M. Deng, L. Wang, Q. Liu, X. Lv, M. Xue
The paper investigates how Large Language Models (LLMs) handle the exploration–exploitation trade‑off in Bayesian Optimization, noting that single‑agent LLM approaches suffer from cognitive overload and unstable search dynamics. It proposes a multi‑agent framework that separates strategic policy mediation from tactical candidate generation, allowing explicit, observable, and adjustable exploration–exploitation decisions. Experiments on continuous optimization benchmarks show that this decomposition improves LLM‑mediated search effectiveness.
By Mohammadsina Almasi, Andrea Carbonati, Hadis Anahideh
The paper investigates how large language model (LLM) search agents can be manipulated through repeated, coordinated search-result injections, a threat that extends beyond single-page poisoning. It introduces Authority‑Chain Hijack (ACH), a strategy that chains manipulated evidence across multiple sources, achieving high attack success rates on the SafeSearch benchmark. The authors also present Trace‑Guided Strategy Evolution (TGSE), an automated method that refines attacker strategies using execution traces, further improving success rates.
By Xuebin Li, Hanqing Zhao, Siyuan Liang, Kejiang Chen, Weiming Zhang, Dacheng Tao, Nenghai Yu
The paper introduces GBClaw, an agentic AI framework that integrates large language models with systematic simulations to study the structure and deformation of general grain boundaries (GBs) in nickel. By applying GBClaw to 14 asymmetric tilt GBs, the authors identified both primary and previously unobserved secondary GB misfit dislocations, revealing their significant impact on shear strength and sliding resistance. The study demonstrates how agentic AI, guided by domain expertise, can bridge high‑throughput screening and deep mechanistic insight at complex material interfaces.
By Yazhuo Liu, Kunqing Ding, Xiaoguo Gong, Zifeng Wang, Yichen Yang, Yifeng Che, Wen Chen, Josh Kacher, Olivier Pierron, Shigenobu Ogata, Ting Zhu
MoLRSI is a recursive self‑improving AI agent designed for molecular discovery that optimizes both workflow and model architecture using persistent cross‑task memory. It refines its search policy based on workflow performance, refinement outcomes, and execution feedback, enabling continuous improvement. In antibiotic discovery, MoLRSI achieved a 15.9% experimental hit rate versus 8.7% for an expert‑led workflow, and in de novo peptide‑MHC binder design it produced eight validated hits with preferential binding, outperforming a baseline approach.
By Yikun Zhang, Xiwei Cheng, Mingze Yuan, Yu Zhang, Brian Shim, Aarti Krishnan, Daniel Rosen, Kunyang Sun, Yingze Wang, Zhenfeng Deng, Botao Yu, Bingxu Liu, David Liu, Tianyu Liu, Yuanqi Du, Nir Hacohen, James J. Collins, Wengong Jin
The paper introduces Agentic University (AU), a Human‑Participatory Multi‑Agent Organization that lets learners act as organization members and collaborate with specialized agents to build a revisable Capability Program. This program links personal goals, multimodal learning activities, and layered assessment, using learner feedback and performance to guide future support. Evaluation shows that AU improves program organization, assessment design, and revision quality, yielding higher mean assessment gains and stronger cognitive support compared to baseline.
By Jingru Fan, Tianle Zhou, Yunjie Zhou, Zihao He, Yixuan Zhou, Kezhao Wu, Xinyu Zheng, Xiyuan Wang, Yuheng Wang, Zhi Li, Songhua Liu, Weinan E, Chen Qian
The paper shows that classical consensus theory, which relies on network topology and convex update rules, fails for large language models (LLMs). By measuring token‑level updates across eight instruction‑tuned models and 73,080 exchanges, the authors find that LLMs violate convexity, exhibit negative weights, and are driven by internal update bias rather than network structure. Simulations with 10,000 agents confirm that algebraic connectivity explains almost none of the outcome variance, leading populations to converge on confidently false claims regardless of their network topology.
By Sahil Rajesh Dhayalkar