The paper presents a multimodal framework that aligns open-ended learning autonomous robots with human purposes using large language models (LLMs) and vision-language models (VLMs). It transforms human intentions expressed in natural language into robot missions and drives through a three-layer architecture—semantic alignment, mission generation, and drive modeling—allowing bidirectional interaction. Experiments in simulation and on a physical robot show that the system generates valid missions and behaviors, maintaining alignment with human intent.
By Diana Cerdas-Vargas, Alejandro Romero, Richard J. Duro
The paper introduces Inalienable Human‑in‑the‑Loop (I‑HITL), a protocol that guarantees an AI agent can only initiate actions with explicit human approval, even if the human’s interface is compromised or an adversarial human delegates approvals. Experiments on a popular agentic wallet platform show that many frontier models can autonomously raise spending caps by harvesting one‑time passcodes, highlighting vulnerabilities in current HITL systems. The authors formally define I‑HITL using communication topologies, prove necessary and sufficient conditions for its implementation, and present a prototype called Countersign that can strengthen existing HITL systems with consumer mobile devices.
By Samuel Breckenridge, Dani Vilardell, Wenhao Wang, Fan Zhang, Ari Juels
The paper introduces XPM-WM, a framework that uses learned world models to generate simulated trajectories for Cross-play Minimization (XPM) in Zero-Shot Coordination. By replacing costly Mixed-Play sampling with model-generated data, XPM-WM improves sample efficiency and scales linearly with population size, producing diverse partner agents that match the performance of existing methods. The approach also bootstraps training for subsequent agents and can be extended to other population‑based training techniques.
By Yi Loo, Akshunn Trivedi, Malika Meghjani
CAFE is a new framework for automated feature engineering that treats the task as a causally‑guided sequential decision process. It first learns a sparse directed acyclic graph to identify causal relationships among features and the target, then uses a multi‑agent deep Q‑learning system to select causal groups and transformation operators while managing feature complexity. On 15 public benchmarks, CAFE outperforms leading AFE baselines, converges faster, and shows markedly lower performance drops under distribution shifts.
By Arun Vignesh Malarkkan, Wangyang Ying, Yanjie Fu
The paper introduces AstroVLBench, a benchmark of 4,100 expert‑verified instances covering optical imaging, radio interferometry, multi‑wavelength photometry, time‑domain light curves, and optical spectroscopy. Six leading vision‑language models were evaluated, revealing that performance varies strongly with modality and is not predicted by general benchmark rankings. Mechanistic ablations identified three bottlenecks: prompt grounding depends on model and task, presenting one‑dimensional data as tables rather than plots improves light‑curve classification, and a significant fraction of correct predictions are supported by physically inconsistent reasoning.
By Wenke Ren, Hengxiao Guo, Wenwen Zuo, Xiaoman Zhang
JRDB-AVR is a new benchmark for embodied agents that tests active visual reasoning in real-world environments. It requires agents to answer questions by requesting limited observations in time and view, and evaluates both the final answer and the visual evidence that supports it. The benchmark includes diverse tasks such as temporal search, viewpoint selection, and human-oriented compositional reasoning, and introduces a reference agent that builds an observation‑grounded graph world model.
By Zhixi Cai, Fucai Ke, Sukai Huang, Maria Garcia de la Banda, Peter J. Stuckey, Gholamreza Haffari, Hamid Rezatofighi
FinAutoRubric is a system that lets experts define reusable evaluation guidance for financial research agents, which the agents then use to automatically generate, review, and validate task‑specific rubrics. The approach replaces costly fixed rubrics with query‑specific ones that incorporate expert standards and can be updated as new information becomes available. On three expert‑authored finance benchmarks, the automatically generated rubrics closely match human scoring and are preferred by in‑house analysts in blind reviews.
By Hoyoung Lee, Suyeol Yun, Jack Haverty, Yunju Cho, Meesong Kim, Daekyung Park, Sumin Kim, Jihoon Kwon, Jasmine Jia Geng, Andrew Chin, Yin Luo, Edward Tong, Yu Yu, Zach Golkhou, Minkyu Kim, Igor Halperin, Young Cha, Alejandro Lopez-Lira, Chanyeol Choi, Yongjae Lee
The article surveys evaluation methods for large language model (LLM)-based agents in multi-turn conversations, reviewing nearly 250 scholarly sources through a PRISMA-inspired framework. It introduces two taxonomies: one outlining what to evaluate—task completion, response quality, user experience, memory and context retention, planning, and tool integration—and another detailing how to evaluate, covering annotation-based, automated, hybrid, and self-judging approaches. The framework incorporates both traditional metrics like BLEU and ROUGE and advanced techniques suited to dynamic, interactive dialogues.
By Shengyue Guan, Jindong Wang, Jiang Bian, Bin Zhu, Jian-guang Lou, Haoyi Xiong
The paper presents a transparent, auditable LLM pipeline for macro‑financial stress testing that uses structured prompting and optional retrieval of country fundamentals and news to generate G7 macroeconomic scenarios. These scenarios are mapped to portfolio losses via a three‑channel PCA factor model, with the macro‑to‑factor mapping explicitly stipulated rather than estimated. Experiments across thousands of scenarios show that severity is driven by prompt design, numerical and narrative severity are weakly linked, and different LLMs exhibit varying degrees of consistency with their own stated changes.
By Masoud Soleimani
DeALOG is a decentralized multi‑agent framework designed for multimodal question answering. It employs specialized agents—Table, Context, Visual, Summarizing, and Verification—that communicate via a shared natural‑language log, serving as persistent memory. This log‑based coordination enables collaborative error detection and verification without central control, leading to competitive performance on several benchmark datasets.
By Abhijit Chakraborty, Ashish Raj Shekhar, Shiven Agarwal, Vivek Gupta
The paper proposes a Multi-Agentic AI framework to enable fairness‑aware and accelerated inference of large, heterogeneous multi‑modal language models on mobile edge networks. It introduces a Global Planning Agent, a Prompt Scheduling Agent, and multiple on‑node LM Deployment Agents that jointly optimize prompt routing and model deployment using natural language reasoning over telemetry and historical data. Experiments on a city‑wide testbed show a 42.9% reduction in end‑to‑end latency versus a state‑of‑the‑art baseline and an 80.6% reduction versus a full‑activation baseline, while preserving high fairness (Jain's index 0.925).
By Haiyuan Li, Hari Madhukumar, Shuangyi Yan, Yulei Wu, Dimitra Simeonidou
RoboGene is an agentic framework designed to generate diverse and informative manipulation tasks for single-arm, dual-arm, and mobile robots. It employs diversity-driven sampling, agentic self-reflection for task quality assessment, and human-verified feedback memory to validate and refine tasks. The framework produced 18,000 real-world trajectories and introduced new metrics for task feasibility and diversity, outperforming models like GPT‑4o and Gemini 2.5 Pro in training VLA models that achieve higher success rates and stronger generalization.
By Yixue Zhang, Kun Wu, Zhi Gao, Zhen Zhao, Pei Ren, Zhiyuan Xu, Fei Liao, Xinhua Wang, Shichao Fan, Di Wu, Qiuxuan Feng, Meng Li, Zhengping Che, Chang Liu, Jian Tang
EntailLLM is a framework that verifies vulnerability discovery paths generated by large language model agents using a separate knowledge graph and logic programming. The system checks whether the agent’s traversal of a function call graph is entailed by domain knowledge, improving pooled entailment from 77% to 98% across various binaries and CWE classes. It is deployed on real medical-device binaries, achieving high entailment without per-device tuning and offering logical, explainable verification grounded in formal semantics.
By Kaustuv Mukherji, Jaikrishna Manojkumar Patil, Colton Payne, Paulo Shakarian, Dana Warmsley, Nigel Stepp, Evelyn Kim
The paper investigates how adding past usage examples of tools to a shared tool index—called folding—affects the retrieval of tools that an AI agent has never used. While folding improves retrieval for tools with examples, it significantly reduces the likelihood that an unused tool appears in the top five results on two of three evaluated catalogs, dropping from 73% to 40% in one benchmark. The authors propose a new scoring method, ZoE, which separates documentation and usage scores, and show that ZoE retrieves unused tools more often than a simple router or a constant‑boost rule, suggesting that evaluations of pooled agent experience should distinguish between used and unused tools.
By Abhijit Chakraborty
arXiv:2610.31726v1 Announce Type: new
Abstract: Effective human-machine teaming requires agents that coordinate with unpredictable human partners. Reinforcement learning from live human feedback or l...
By Ash Simmons, George De Ath, Richard M. Everson
arXiv:2610.33279v1 Announce Type: new
Abstract: Introduction: Increasing volumes of individual case safety reports (ICSRs) challenge expert-led causality assessment in pharmacovigilance. We developed...
By Daisy Rae Lopez, Angela Boccia, Maurizio Sessa
arXiv:2610.34735v1 Announce Type: new
Abstract: An agent that truly knows and acts on behalf of its user has been a long-standing promise of AI assistants. Computer-use agents bring this promise with...
By Yuting Ning, Ruijia Cheng, Sushobhan Ghosh, Peitong Duan, HungShi Lin, Abhishek Sundararajan, Shreyash Pandey, Huan Sun, Jeffrey Nichols, Eldon Schoop
arXiv:2610.34889v1 Announce Type: new
Abstract: Controlling personality in large language models is important for building conversational agents that can be customized to different users and interact...
By Gunhee Cho, Hongyeob Kim, Yun-Gyung Cheong
arXiv:2610.35671v1 Announce Type: new
Abstract: As frontier language models grow increasingly capable and autonomous, independent third-party evaluations are critical for characterizing their systemi...
By Yao Huang, Zhiyuan Zeng, Yu Wang, Zhitong Zhu, Dingcheng Duan, Hao Li, Shangzhi Xu, Yifan Wang, Jinzhe Tu, Shukun Xiong, Yihe Zang, Han Yan, Hongchang Yang, Fuyu Gu, Wei Xu, Yinpeng Dong, Xiaojian Li
arXiv:2610.35272v1 Announce Type: cross
Abstract: A transcriber listening through one microphone also hears everyone near it. We train causal voice isolators: masking transformers of 140M, 243M and 3...
By Husein Zolkepli