The paper introduces a prediction-informed hierarchical control framework for autonomous UAV landing on a USV in high sea states. It uses a pretrained model to explicitly predict USV attitude, enabling the identification of suitable landing windows and optimal touchdown timing. A conventional low‑level flight controller then maintains UAV stability while following the motion commands.
By Kaizhen Li, Xi Zhou, Dan Zhang, Xiaowei Li
Fly-by-Code introduces embodied coding agents that generate robot policies from natural-language instructions and refine them using execution feedback. In aerial manipulation, partial observability and fragile interactions limit the usefulness of standard feedback, so the authors propose Active Feedback (AF), a framework that synthesizes diagnostic programs to view unseen outcomes or probe failure causes. Experiments on six simulated tasks show AF dramatically reduces false completions and boosts task success, and the refined programs transfer to a real aerial manipulator without code changes.
By Jaewoo Lee, Jeongyeon Seo, Sihyun Cho, Gyeongrak Choe, Yutong Wang, Bavin Saravanan, Jia-Bin Huang, Furong Huang, Sebastian Scherer, Guanya Shi, H. Jin Kim, Seungjae Lee, Dongjae Lee
The paper introduces MisVisAgentBench, a paired benchmark with 280 task instances that compare clean versus misleading chart evidence for web agents. It evaluates how deceptive visualizations affect agent execution, providing a diagnostic framework to separate misleading-induced failures from baseline difficulty. Results show that misleading charts reduce task success by over 34 percentage points and tend to redirect agents toward targeted misleading actions.
By Yisheng Li, Dongming Han, Hongkun Pan, Bingcheng Mao, Zijian Wang, Yunlong Mao, Wei Tong, Sheng Zhong
The paper introduces TerraAgent, a training‑free framework that enhances remote sensing visual question answering by integrating three modules: GDR for demonstration retrieval, MACE for evidence‑typed sub‑question decomposition, and CVA for counterfactual evidence auditing. It demonstrates significant improvements on multiple benchmarks, raising accuracy, visual‑dependency scores, and reducing prior‑only answers and hallucinations without fine‑tuning.
By Jingwen Huang, Yiming Fang, Hanxiao Wu, Boyuan He, Keyi Luo
The paper introduces SAMF, a Structure-Aware Matching Framework designed to improve UAV thermal-to-satellite geo‑localization by exploiting stable cross‑modal structural cues. It comprises a Directional Structure Enhancement Module that captures directional and low‑frequency spatial patterns, and a Modality‑Aware Affine Remapping Module that aligns these enhanced features across modalities. Experiments on Boson‑nighttime and Thermal‑UAV‑Paired datasets demonstrate state‑of‑the‑art performance, with ablation studies confirming the complementary benefits of the two modules.
By Chun Liu, Yuxin Wang, Zhenxiang Huo, Linlin Zhu, Tao Xu, Henggang Han
MeshBind introduces an automatic annotation pipeline that generates per-point scene flow for non‑rigid objects such as pedestrians by reconstructing articulated mesh models from six surround cameras and registering them to LiDAR data. The method employs a local‑frame binding technique that attaches each point to a mesh via a per‑triangle tangent‑normal frame, and a three‑layer flow registration pipeline that compensates ego‑vehicle motion, performs box binding, and applies mesh binding to produce full‑scene flow. MeshBind is implemented in the SF‑Studio tool, and the authors release the SUScape‑Flow Benchmark (SFB), a large‑scale dataset with 41k frames of per‑point non‑rigid pedestrian annotations and a comprehensive evaluation protocol, demonstrating that fine‑tuning on SFB reduces endpoint error by 2‑3× and equalizes non‑rigid and rigid EPE.
By Yinqi Chen, E Li, Shuaijun Wang, Weiyuan Peng, Meiying Zhang, Qi Hao
ConClick introduces a contrastive learning framework that improves GUI grounding by teaching models to distinguish target interface elements from plausible distractors. The authors built ConClick-Data, a dataset of 4,292 challenging target-distractor pairs, and trained the model with listwise ranking and pairwise margin objectives, achieving significant accuracy gains on multiple benchmarks. An inference-time strategy called ReClick further boosts performance by revisiting promising regions and comparing competing predictions.
By Ruotian Zhang, Yuhui Zhang, Junzhe Huang, Yu Xin, Hehe Fan, Xiaoxiao Sun
The paper examines why latent action models (LAMs) struggle when training on unlabeled videos that contain exogenous state such as background clutter. By extending a linear LAM framework to explicitly model exogenous state, the authors analytically show that the standard reconstruction objective causes latent actions to encode future exogenous information, and that focusing the representation on endogenous components mitigates this interference. They also demonstrate that auxiliary objectives like action-supervision encourage consistency across exogenous states, and validate these insights with experiments on both linear and nonlinear LAMs.
By Jung Min Lee, Taehyun Cho, Li Zhao, Jungwoo Lee
CLEAT (CLarification through valuE-guided Agent co-Training) is a framework that jointly trains a clarification-value controller and a task policy to improve interactive intent alignment in large language model agents. It introduces a residual query value metric, estimated via paired rollouts, to assess the utility of asking clarification questions and uses this to guide both controller predictions and policy decisions. Experiments with Qwen3-4B show CLEAT outperforming baselines across multiple benchmarks, with significant gains in user‑centric tasks and demonstrating that co‑training benefits the task policy even when the controller is omitted at inference.
By Jiatan Huang, Zheyuan Zhang, Ziming Li, Tianyi Ma, Mingchen Li, Yanfang Ye, Chuxu Zhang
VEPA is a framework that enables the verified evolution of formal guardrails for lifelong large language model agents. It represents guardrails as formal policies of symbolic rules over typed agent events, and during deployment uses LLM‑guided feedback to propose policy patches that fix new failures while preserving correct behavior. A deterministic verifier accepts only structurally valid, conflict‑free policies, and an evidence‑aware authority layer activates verified rules only when sufficient supporting evidence exists, ensuring that uncertain updates do not directly affect guardrail decisions.
By Duy Nguyen, Lesly Miculicich, Bhavana Dalvi Mishra, Shuo Li, Krishnamurthy Dj Dvijotham, Mohit Bansal, Tomas Pfister, Long T. Le
SmartSalesBench is a new multi‑turn conversational benchmark designed to evaluate large language model sales agents against buyer personas that are explicitly programmed to resist and challenge the agents. The benchmark features five agent profiles, twelve resistant buyer personas, fifteen smartphone products across six price tiers, and includes 13,680 sessions across four controlled experiments. Evaluation uses nine metrics grouped into primary, secondary, and exploratory tiers, combining automated LLM‑as‑judge scoring with human assessment.
By P N Bhargav Teja, Velpula Lokesh Reddy, Madhu Basumata, Susen Sarkar, Rajasekhara Reddy Duvvuru Muni
Work2Skill is a framework that captures human GUI experience to induce reusable agent skills. It introduces WorkRecorder, an infrastructure that aligns screen recordings, mouse and keyboard events, and accessibility-tree data, and uses it to build WorkDemo-350, the first synchronized multimodal dataset of 350 human desktop GUI demonstrations. The proposed skill‑induction method extracts workflow rules, step dependencies, and completion checks, achieving higher scores and success rates than baseline methods on MyPCBench and OSWorld benchmarks.
By Yuqi Huang, Hui Li, Yifei Cao, Zikun Cai, Tianci Xie, Yangfan Gao, Pinxue Guo, Chaoyun Zhang, Bingxuan Li, Rui Zheng, Hang Yan, Ning Liao
The paper introduces Hindsight Structured Advice Distillation (HSAD), a method that trains a small advisor model to generate structured guidance for a frozen executor language‑model agent. By analyzing executor failures on a training set, a teacher writes advice that would have helped, and the advisor learns to produce this advice from input alone, adding minimal latency. Experiments on the FrontierFinance benchmark show that an 8B advisor boosts the qualification rate of Samaya’s research agent from 52.9% to 56.6%, surpassing the agent’s high‑effort configuration while cutting cost per query by 37%, and the approach also improves performance on WebShop, BIRD text‑to‑SQL, and ExpertLongBench.
By O. Ozan Koyluoglu
The study evaluates the impact of integrating the Jev decision model into five AI agents across two application domains—embodied tasks in ALFWorld and mobile app use in AndroidWorld. While Jev consistently lowers decision cost and latency, the resulting task-level benefits differ by workflow: task success remains near baseline in ALFWorld but drops noticeably in two mobile agents. The authors find that auxiliary reasoning, additional interactions, failed recoveries, and premature completions can offset decision savings, leading to higher cost per successful outcome in some cases.
By Yuchen Ling, Shengcheng Yu, Zhenyu Chen, Chunrong Fang
KnowForge is a modular pipeline for open‑domain multimodal document question answering that separates retrieval, judging, and reading into distinct agents. It uses a hybrid‑retrieval knowledge‑point index, a Locator agent for searching, a Bridge Judge to enforce a second hop, a Deep Reader that handles text, images, and tables, and a synthesizer with format checking. On the M3DocVQA benchmark it achieves an F1 of 47.5, outperforming prior systems by 11 points, and shows strong gains on MultiHop‑RAG and MMLongBench‑Doc as well.
By Wenlin Lin, Fanzhe Wei
RoleWorld is a symbolic world‑modeling framework that uses reusable prototypes and environment‑specific parameters—called functional role abstractions—to capture how objects behave in interactions. A teacher LLM extracts role prototypes and executable transition rules from interaction trajectories, while an actor LLM employs a skill to apply these roles during decision making. In target environments, the actor refines role assignments and parameter estimates online through active testing, without updating LLM weights, and experiments on the MARS benchmark demonstrate robust generalization across environments with altered object behavior.
By Boyuan Xiao, Pu Wang, Ji Feng, Yao-Xiang Ding
The paper presents a local agentic AI framework that uses laptop-based language models to autonomously generate and refine edge case scenarios for autonomous systems. It employs a think‑act‑observe‑update loop with JSON encoding, contextual prompts from system requirements and environment descriptions, and a second LM that critiques scenarios against domain criteria. The refined scenarios are simulated, evaluated, and clustered to identify functional gaps, providing traceable evidence for safety assurance.
By Victoria J. Hodge, Colin Paterson, Rob Alexander
The paper introduces MENTOR, a multi‑agent framework that provides on‑policy pedagogical intervention to address zero‑reward collapse in reinforcement learning from verifiable rewards. MENTOR diagnoses failures in real time and generates targeted prompts to guide the learning agent, integrating seamlessly with existing policy optimization pipelines such as GRPO. Experiments on math and code generation tasks show that MENTOR significantly reduces collapse, achieving up to 78.3% relative gains on Llama‑3.2‑3B‑Instruct.
By Guanlan Yue, Hang Yu, Xiaoyong Huang, Ruipeng Li, Wei Zhang, Bingchang Liu, Rui Wang
The paper introduces HOA (Humanize Olympiad Agent), a workflow that orchestrates multiple LLM agent sessions via reviewer feedback, a structured wiki, and a verifier to tackle olympiad problems. Using GPT‑5.6 Sol, HOA achieves perfect scores on the IMO, IPhO, and IOI 2026, near‑full marks on the IChO 2026, and solves all 672 PutnamBench problems, while ranking first on the Lean‑Eval leaderboard. With open models like Kimi‑K3, HOA still attains full scores on major olympiads and a gold‑level IChO score, showing that systematic feedback can close gaps between closed‑source and open‑source agents.
By Zhengyang Zhang, Jing Xiong, Hongzhou Lin, Juihui Chung, Zijian Zhang, Song Han, Sihao Liu, Ligeng Zhu
ReHarness is a recurrent harness framework designed for agentic continual learning, enabling language model agents to adapt online to distribution shifts. It augments forward read/write operations with backward revision, using feedback from each instance’s trajectory to refine the harness state. Experiments on three continual learning benchmarks and multiple language model backbones show that ReHarness outperforms baseline methods, producing source‑grounded revisions that improve stability and plasticity.
By Yuxin Xiao, Zhiyun Lu, Jianyu Wang, Yangtian Zhang, Yang Song, Qifan Wang, Bailin Wang