Pai 的 AI 雷达

今日必看

今天的简报围绕“智能体如何被构建、接入、评估并落到具体工程”展开：既有知识库与可视化工作流、端到端评估、自主建模和开发工具，也有 AIGC 视频检测、C++ 代码智能、ROS 2 加速与机器人场景。阅读时把 featured 视为今日主线，再用 why_items 中的动作、接口和约束提示，把条目转成可验证的工作步骤。

日期
2026/09/23
今日条目
85
往期2026/09/23
全部85编码代理2Agent与工具6个人助理3模型与评测33创作成片13开发与开源11产品接入17
领域机器人具身3生信医疗4金融1安全治理1行业观察8创业者1

更多资讯

#26

SLICEChat: Progressive In-Encoder Token Pruning for Whole-Slide Pathology Language Models

arXiv:2609.24894v1 Announce Type: new Abstract: Whole-slide pathology images (WSIs) contain gigapixel-scale visual content, creating a major scalability challenge for slide-level multimodal large language models (MLLMs). Existing approaches process thousands of patch tokens and t

推荐理由面对 gigapixel 级全切片病理图像，可在编码器内部逐步裁剪 token，先控制输入规模，再观察病理语言模型的感知与文本表现是否保持；评测同时记录 token 数、可扩展性和输出质量。
模型与评测arXiv cs.CV 官方
arxiv.org打开原文
#27

Grow the Harness, Not the Context: From Strategy-Free Scaffolds to Reusable Specialist Agents

arXiv:2609.26760v1 Announce Type: new Abstract: Large language model (LLM) agents often handle streams of related tasks, yet standard harnesses repeatedly ask the model to reconstruct the same control decisions inside each task's context. We study whether task feedback can instea

推荐理由可把重复任务中的控制决策从每次上下文里抽出来，沉淀进 harness，并用任务反馈逐步培养可复用的 specialist agent；这样扩展的是控制骨架而非不断加长上下文，需持续记录反馈如何改变决策。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#28

AgriGen: Large-Scale Scene Generation Framework for Photorealistic Agricultural Robotics Simulation

arXiv:2609.25725v1 Announce Type: new Abstract: Agricultural robotics is advancing rapidly, yet progress remains constrained by limited field access, lack of control over field conditions, geographic variability, and seasonal crop cycles. These factors make it difficult and costl

推荐理由可用 AgriGen 为农业机器人仿真生成大规模、写实场景，把真实田间难以控制的地理差异、季节周期和进入条件转成可重复的模拟变量；再用这些场景检验机器人策略在不同农田条件下的泛化。
模型与评测arXiv cs.RO 官方
arxiv.org打开原文
#29

Design and Implementation of an Ultra-Low-Cost Wall-Climbing Robot for Infrastructure Crack Detection

arXiv:2609.26130v1 Announce Type: new Abstract: Crack detection is a crucial process to ensure the safety and longevity of buildings and other infrastructure. In this paper, we developed a low-cost, automated crack detection robot that leverages CNN, EfficientNet-B0, and YOLOv8 f

推荐理由可按该方案搭建基础设施裂缝检测链路：让低成本爬墙机器人采集墙面图像，再用 CNN、EfficientNet-B0 与 YOLOv8 组合识别裂缝；验证时同时记录攀爬覆盖和检测结果，避免只看模型精度。
模型与评测arXiv cs.RO 官方
arxiv.org打开原文
#30

AgenticSizing: A Large Language Model-based Multi-Agent Framework for Analog Circuit Sizing

arXiv:2609.25873v1 Announce Type: new Abstract: Analog circuit sizing remains a challenging and time-consuming task due to the large design space, strong performance trade-offs, and increasing circuit complexity in scaled technologies. Although recent large language model (LLM)-b

推荐理由可把 AgenticSizing 作为模拟电路 sizing 的多智能体框架入口：将大设计空间和性能权衡拆给协作 agent 处理，再用统一指标核对候选方案；流程设计要保留对复杂度和技术尺度的约束，避免只追求单点结果。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#31

CQ4OE: A benchmark for assessing LLM-assisted ontology generation from competency questions

arXiv:2609.26029v1 Announce Type: new Abstract: Ontology generation from Competency Questions (CQs) is a central yet labor-intensive phase of Ontology Engineering. While large language models (LLMs) offer promising automation capabilities, current evaluations remain fragmented. T

推荐理由可用 CQ4OE 把 competency questions 驱动的本体生成做成可测基准：先固定问题集，再比较 LLM 生成的本体结果，并用统一维度补齐目前分散的评估；这样能把生成流程从示范性输出推进到可复现比较。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#32

From Experts to Sub-experts: Fine-grained Parameter-Efficient Fine-Tuning for MoE LLMs

arXiv:2609.25655v1 Announce Type: cross Abstract: As large language models (LLMs) scale rapidly, dense full-parameter adaptation becomes increasingly expensive, motivating sparse and modular architectures such as Mixture-of-Experts (MoE) models. This shift raises a key question f

推荐理由在 MoE LLM 适配流程中，可将全参数训练替换为更细粒度的参数高效微调，把更新范围压到需要的专家或模块；评估时同时记录适配成本与效果，保留稀疏、模块化结构带来的取舍。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#33

Passes Alone, Fails Together: Benchmarking Semantic Coordination in Parallel LLM-Agent Development

arXiv:2609.25396v1 Announce Type: cross Abstract: Parallel coding agents can produce patches that work alone but fail when merged. This happens when one agent changes an interface or rule that another agent still relies on. We study these failures with stale, a benchmark for sema

推荐理由并行 coding agent 协作时，应把合并后的整体行为纳入测试：逐项检查一个 agent 改动的接口或规则是否仍被其他 agent 依赖，并记录“单独通过、合并失败”的语义协调案例；接口变更要同步传播。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#34

Quantifying Overclaiming Propensity in Frontier LLM Agents

arXiv:2609.20812v3 Announce Type: replace-cross Abstract: Frontier coding agents are increasingly trusted to work autonomously for long periods of time, yet what they actually did is often hard to tell from their final response. We quantify the propensity of such agents to overcl

推荐理由可把 Quantifying Overclaiming Propensity 的思路用于 agent 审计：把最终回答与实际执行记录逐项核对，标出声称完成但证据不足的步骤，再统计长期自主运行中的过度声称倾向；验收不能只看总结措辞。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#35

Qwen-Audio-3.1-Realtime: Towards Reliable Agentic Voice Interaction

arXiv:2609.25176v1 Announce Type: cross Abstract: Real-time voice assistants must reason over evolving requests, execute actions, and follow conversational rules. Qwen-Audio-3.1-Realtime brings these requirements together through Think, Act, and Speak and Coordinate. Think combin

推荐理由构建实时语音 agent 时，可按 Qwen-Audio-3.1-Realtime 的 Think、Act、Speak、Coordinate 拆分流程：先处理变化中的请求，再执行动作并维持对话规则，同时检查各阶段的衔接；评测不能只看语音输出。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#36

QwenVLConnector: A Fast, Unified Medical VLM Chatbot for Fine-Grained Clinical Perception and Text Generation

arXiv:2609.23139v1 Announce Type: new Abstract: Most medical vision-language models (VLMs) excel at open-ended report generation and VQA but provide limited support for structured, fine-grained clinical perception within a unified interface. We present QwenVLConnector, a Qwen2.5

推荐理由可用 QwenVLConnector 设计一个统一的医疗 VLM 入口，把细粒度临床感知与文本生成放进同一交互流程；验收时分别检查结构化感知、开放式报告和 VQA，避免只用报告生成结果代表全部能力。
模型与评测生信医疗arXiv cs.CV 官方
arxiv.org打开原文
#37

ReasonLab: A Controlled and Auditable Evaluation of Prompting Techniques for Multiple-Choice QA

arXiv:2607.14109v2 Announce Type: replace-cross Abstract: Probing the capabilities of Large Language Models (LLMs) and building robust solutions for Multiple-Choice Question Answering (MCQA) remain central challenges in natural language understanding. Furthermore, the rapid proli

推荐理由可用 ReasonLab 搭建受控、可审计的多选题问答评测：固定 prompting techniques，逐项记录模型答案与评测条件，再比较不同提示策略对 MCQA 能力和稳健性的影响；结果要能复核。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#38

Recovering Agentic Sovereignty: Mitigating the Consensus Paradox via Contrastive Epistemic Decoding

arXiv:2609.25570v1 Announce Type: new Abstract: Large language models (LLMs) exhibit a parametric vulnerability to adversarial swarm consensus. To mitigate this sycophancy, we introduce Contrastive Epistemic Decoding (CED), a zero-shot inference intervention. Unlike standard Cont

推荐理由在面对对抗性群体共识或迎合风险时，可把 Contrastive Epistemic Decoding 作为零样本推理干预加入测试，比较标准解码与 CED 的回答差异；把是否受共识牵引设为可靠性检查项。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#39

Reducing Hallucinations in Large Language Models Through Integrated Self-Verification and Retrieval-Augmented Generation

arXiv:2609.26229v1 Announce Type: cross Abstract: Large Language Models (LLMs) are progressively used for advanced engineering tasks, includes Computer-Aided Design (CAD) documentation, standards compliance verification, and knowledge retrieval. Still, they are prone to produce h

推荐理由处理 CAD 文档、标准合规核验或知识检索时，可把 RAG 与自验证串成两步：先从资料检索依据，再让模型复核生成内容与依据的一致性；把幻觉率和核验通过情况纳入验收，而非只看回答是否流畅。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#40

Simulate to Generalize: Scaling Stateful Supervision for API-calling Agents using LLM World Models

arXiv:2607.16900v3 Announce Type: replace Abstract: Training agents that generalize to unseen, stateful environments requires a massive dataset of state-changing trajectories covering a vast and diverse set of APIs. However, scaling this broad supervision is severely bottlenecked

推荐理由训练会调用 API 的 agent 时，可用 LLM world model 模拟状态变化轨迹，扩充覆盖不同 API 和未见环境的监督；流程里要保留状态转移与动作结果，检查模拟数据能否支持真正的泛化，而非只增加样本量。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#41

Spectra: A Rules-Driven LLM Pipeline for Automated KYC Document Processing

arXiv:2609.25474v1 Announce Type: new Abstract: Know Your Client (KYC) onboarding in capital markets requires analysts to manually classify documents, extract structured data from heterogeneous sources, and validate compliance against complex regulatory policies. This process req

推荐理由可把 KYC 入职拆成规则驱动的 LLM 流水线：先分类异构文件，再抽取结构化字段，最后按复杂监管政策做合规校验；每一步保留规则与结果的对应关系，便于复核自动处理是否可靠。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#42

TREND-10K: A Comprehensive Dataset for Next-Generation Video Quality Assessment Based on Preference-Driven Media

arXiv:2609.26187v1 Announce Type: new Abstract: The increasing prominence of short-video platforms, coupled with the advanced commercialization of AI-generated content (AIGC) videos, has led to a shift in the types of video media trend consumed by users in their daily lives. Trad

推荐理由做下一代视频质量评估时，可用 TREND-10K 作为偏好驱动媒体数据集，覆盖短视频与 AIGC 视频场景；先按用户偏好组织样本，再验证模型对不同视频媒介趋势的质量判断是否一致。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#43

Imperfection for Precision: Upcycling Imperfect Data for High-Precision Robotic Manipulation

arXiv:2609.26672v1 Announce Type: new Abstract: Training vision-language-action (VLA) models for high-precision manipulation typically requires task-specific, high-quality data (e.g., teleoperation), which is slow and expensive to collect. To reduce this burden without compromisi

推荐理由训练高精度操作 VLA 时，可把不完美数据作为可再利用的训练材料，减少对慢且昂贵的高质量遥操作数据依赖；流程中要同时检查数据转换后精度是否保持，确保降低采集成本不牺牲操作表现。
模型与评测arXiv cs.RO 官方
arxiv.org打开原文
#44

Toward User-Mediated Self-Repair in Ubiquitous Robots Through Goal-Oriented Agentic AI

arXiv:2609.26157v1 Announce Type: new Abstract: Ubiquitous robotic systems often lack traditional visual interfaces, necessitating resilient natural language interaction for maintenance and repair tasks. This paper presents a goal oriented agentic AI architecture designed to enab

推荐理由Toward User-Mediated Self-Repair：arXiv:2609.26157v1 Announce Type: new Abstract: Ubiquitous robotic systems often lack。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#45

3D Gait-Based Autism Classification Using Attention-Enhanced Deep Learning with Cross-Fold Statistical Stability Analysis

arXiv:2609.14159v2 Announce Type: replace Abstract: Autism Spectrum Disorder (ASD) is a neurodevelopmental condition whose early diagnosis remains challenging because conventional clinical assessments are often subjective, time-consuming, and require expert evaluation. Gait provi

推荐理由3D Gait-Based Autism Classification：arXiv:2609.14159v2 Announce Type: replace Abstract: Autism Spectrum Disorder (ASD) is。
模型与评测生信医疗arXiv cs.CV 官方
arxiv.org打开原文
#46

A Hybrid AI Framework for Academic Advising: Integrating Ensemble-Based Grade Prediction and a Rule-Based Expert System

arXiv:2609.26243v1 Announce Type: new Abstract: The rapidly increasing student population has posed serious challenges to the traditional academic advising process. This study designs and implements a multi-purpose intelligent system to support students' academic progress, based

推荐理由A Hybrid AI Framework for Academic：arXiv:2609.26243v1 Announce Type: new Abstract: The rapidly increasing student。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#47

A2M: Trace-Optimized Agent Hijacking in the MCP Ecosystem

arXiv:2609.26761v1 Announce Type: cross Abstract: Agents using the Model Context Protocol (MCP) rely on semantic matching to select tools from third-party servers, exposing a semantic supply-chain risk through attacker-controlled metadata and outputs. We introduce A2M (Attraction

推荐理由A2M: Trace-Optimized Agent Hijacking：arXiv:2609.26761v1 Announce Type: cross Abstract: Agents using the Model Context。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#48

AgentSTAR: Agentic Shape Tracking and Reconstruction from Monocular Videos

arXiv:2609.24487v1 Announce Type: new Abstract: In this work, we present a method for shape reconstruction and tracking from video via agentic analysis-by-synthesis. Unlike prior methods which first estimate dense pixel correspondences and then recover object motion from them, ou

推荐理由AgentSTAR: Agentic Shape Tracking：arXiv:2609.24487v1 Announce Type: new Abstract: In this work, we present a method。
模型与评测arXiv cs.CV 官方
arxiv.org打开原文
#49

An Efficient and Effective Watermarking Scheme for the Protection of the Intellectual Property Rights of Video Generative Models

arXiv:2609.23586v1 Announce Type: new Abstract: The rapid development of video generative models (VGMs) has enabled the generation of highly realistic synthetic videos, raising concerns about the intellectual property rights (IPR) of these models. In particular, two closely relat

推荐理由An Efficient and Effective Watermarking：arXiv:2609.23586v1 Announce Type: new Abstract: The rapid development of video。
模型与评测arXiv cs.CV 官方
arxiv.org打开原文