Daily Papers · 2026-10-14
📄 论文日报 | 来源：Hugging Face Daily Papers
🗓 Episode：2026.10.13 | 视觉定位模型领先；编码智能体构建堪忧
1. GroundingPI：面向物理智能的视觉基元定位基础模型
- 英文标题：GroundingPI: A Grounding Foundation Model towards Physical Intelligence with Visual Primitives
- arXiv：2609.39601
💡 一句话解读：机器人等物理智能系统需要快速、精准地找到目标物体，但现有通用视觉模型在杂乱或微小物体上常失败。本文提出GroundingPI，一个40亿参数的定位基础模型，用统一坐标输出点和框，结合多模态预训练、微调和强化学习，在34个基准上超越更大模型，并提升操作与驾驶任务表现。
精准定位至关重要：它要指明哪个物体是目标、该物体在哪里，即使在杂乱场景和微小物体上也要做到，并且必须足够快以支持闭环控制。然而，视觉-语言-动作（VLA）模型和世界-动作模型（WAMs）从通用视觉-语言和视频生成骨干网络中获取感知能力，在这些场景下仍然表现不佳。我们提出GroundingPI，一个4B参数的定位基础模型，以共享词汇表中的量化坐标生成点和框。其训练结合了多模态与空间预训练、监督微调以及使用GRPO的强化学习，并利用来自公开数据集和专用数据引擎的监督。在涵盖11种感知能力的34个定位基准上，与44个基线相比，GroundingPI确立了新的最优性能，平均达到73.68%，高于规模更大的GPT-6 Astra（71.54%）。作为下游视觉骨干，GroundingPI提升了机器人操作和自动驾驶的性能。在RoboTwin 2.0上，它在所有四种分布外设置中均优于我们评估的每个主流骨干，相对最强骨干最高提升24.8%。在RoboCasa-GR1上，仅用50%演示数据训练的GroundingPI优于用75%数据训练的基线。在nuScenes上，作为视觉骨干，GroundingPI实现了0.296米的平均开环L2误差。我们系统分析了GroundingPI在规模和數據组成上的预训练。下游自动驾驶和机器人操作随着预训练规模扩大而提升。分析这11种感知能力的数据配方表明，密集定位对两者都有显著益处，而OCR有潜力成为感知学习的催化剂。这些结果支持将定位作为感知基础，并将专用感知预训练作为物理智能基础模型的一个有前景的方向。
📄 English Abstract
Precise grounding matters. It specifies which object is the target and where that object is, even in clutter and for tiny objects, and it has to be fast enough for closed-loop control. Yet vision-language-action (VLA) and world-action models (WAMs) take perception from general-purpose vision-language and video-generation backbones, which still fail in these settings. We introduce GroundingPI, a 4B grounding foundation model that generates points and boxes as quantized coordinates in a shared vocabulary. Training combines multimodal and spatial pretraining, supervised fine-tuning, and reinforcement learning with GRPO, using supervision from public datasets and dedicated data engines. Against 44 baselines across 34 grounding benchmarks spanning 11 perceptual capabilities, GroundingPI establishes a new state of the art, averaging 73.68%, above the larger GPT-6 Astra (71.54%). As a downstream visual backbone, GroundingPI improves performance on robotic manipulation and autonomous driving. On RoboTwin 2.0, it outperforms every mainstream backbone we evaluate in all four out-of-distribution settings, by up to 24.8% relative to the strongest backbone. On RoboCasa-GR1, GroundingPI trained with 50% of the demonstrations outperforms those baselines trained with 75%. On nuScenes, used as the visual backbone, GroundingPI attains an average open-loop L2 error of 0.296 m. We systematically analyze GroundingPI's pretraining in scale and data composition. Downstream autonomous driving and robotic manipulation improve as the pretraining is scaled. Analyzing the data recipe across these 11 perceptual capabilities shows dense grounding's substantial benefits for both, and OCR's potential as a catalyst for perceptual learning. These results support grounding as a perceptual foundation, and dedicated perceptual pretraining as a promising direction for foundation models of physical intelligence.
2. AgentSWE：编码智能体能否构建你真正想要的智能体？
- 英文标题：AgentSWE: Can Coding Agents Build the Agent You Actually Want?
- arXiv：2610.13361
💡 一句话解读：这篇研究要解决的是：当AI编程助手帮人构建智能体时，它交付的智能体是否真能满足需求，而非只是表面像。作者提出AgentSWE基准，用25个任务分创建、编辑、优化三阶段测试，发现现有编码智能体普遍存在过拟合、空壳运行、虚假完成等问题。
随着软件生产逐渐交由软件自身完成，编码智能体（coding agents）越来越多地被委托去构建适配用户自身工作流的智能体。这种委托隐含一个不易察觉的假设：交付的智能体将忠实服务于其被构建时所针对的需求，而非仅仅表面上如此。目前尚无现有基准测试检验这一假设。我们提出AgentSWE，一个涵盖智能体开发生命周期三个阶段的智能体软件工程基准：根据自然语言需求创建完整智能体、编辑生产级智能体代码库、以及面向可度量目标进行优化。其25个任务均以“委托”而非“考试”的形式设定：构建者针对少量示例用例进行迭代，而验收则在设计覆盖完整需求的留出用例上运行冻结的提交，依据需求推导出的评分标准在硬性通过/失败门槛下评分，或在优化任务中依据源基准自身的客观指标评分。在跨模型和工具链评估多种编码智能体后，我们发现没有一个能接近满足需求所要求的结果：构建者所演示的能力与其实际交付之间的差距是常态，而非边缘情况。伴随对失败轨迹的实证分析，我们识别出反复出现的失败模式，包括对可见用例的过拟合、代码看似完整却运行时空洞、以及无依据的完成声明，每一种都是编码智能体可借以学习构建所需智能体的信号。
📄 English Abstract
As the production of software passes into the hands of software itself, coding agents are increasingly commissioned to build agents fitted to their users' own workflows. This delegation carries a quiet assumption, that the delivered agent will faithfully serve the requirement it was built for, rather than merely appear to. No existing benchmark tests that assumption. We introduce AgentSWE, a benchmark for agent software engineering spanning three stages of the agent development lifecycle: Creation of a complete agent from a natural-language requirement, Editing of production agent codebases, and Optimization toward a measurable target. Each of its 25 tasks is framed as a commission rather than an exam: the builder iterates against a handful of illustrative cases, while acceptance exercises the frozen submission on held-out cases designed to span the full requirement, grading behavior against requirement-derived rubrics under hard pass/fail gates or, for Optimization, against the source benchmark's own objective. Evaluating a range of coding agents across models and harnesses, we find that none comes close to delivering what a requirement asks for: the gap between what a builder demonstrates and what it delivers is the norm, not the corner case. An accompanying empirical analysis of the failure trajectories identifies recurring failure modes, among them overfitting to the visible cases, hollow runtimes behind code that looks complete, and unsupported claims of completion, each a signal by which coding agents can learn to build the agents they are asked for.
3. 面向长时程终端智能的智能体数据规模化
- 英文标题：Scaling Agentic Data for Long-Horizon Terminal Intelligence
- arXiv：2610.13166
💡 一句话解读：现有终端智能体训练数据多为短任务，无法体现智能体通过探索积累证据、根据反馈调整策略并产出最终成果的长期行为。为此，作者提出全自动数据合成引擎 TerminalHorizon，从真实职业工作出发，识别并补回被简化掉的关键工作，再逐步扩展任务，构建 1.5K 任务环境，训练出 TerminalHorizon-27B，在 12 项基准上大幅提升。
终端智能体正从本地操作迈向完成真实世界的专业任务。短轨迹可以展示命令和工具调用，但无法充分体现智能体如何通过探索积累证据、据此形成判断、根据反馈持续调整策略，并将中间结果转化为最终交付物。高质量的长时程轨迹能够捕捉这些智能体行为模式及其在单一问题解决过程中的关联，为持续自主工作提供更丰富的经验。然而，环境合成可能提前完成或简化这些模式背后的关键工作，从而消除了智能体本应进行的探索、实现与验证。为解决这一问题，我们提出 TerminalHorizon，一个面向长时程终端智能体的全自动数据合成引擎。它从真实世界专业工作出发，利用独立执行来识别构建过程中被省略的关键工作，并用由此获得的证据指导环境重建。为将这些模式扩展到不同阶段，它在同一目标下逐步扩展任务，引入基于先前结果的新需求，并要求进一步探索与改进。利用该引擎，我们构建了 1.5K 个任务环境，并在所得数据集上对 Qwen3.5-27B 进行监督微调，得到 TerminalHorizon-27B。我们在涵盖编码智能体任务、真实生产环境中的智能体任务以及通用推理的 12 个基准上均取得了显著性能提升。多个 RSI 案例研究（每个拥有 12 小时研究预算）进一步表明，该模型能够超越初始方案继续实验、根据反馈改进方法，并在最终交付物中保留有效的改进。
📄 English Abstract
Terminal agents are moving beyond local operations toward completing real-world professional tasks. Short trajectories can demonstrate commands and tool calls, but they do not fully show how agents accumulate evidence through exploration, use it to inform judgments, continually adjust strategies based on feedback, and turn intermediate results into final deliverables. High-quality long-horizon trajectories capture these agentic behavioral patterns and their connections within a single problem-solving process, providing richer experience for sustained autonomous work. However, environment synthesis may complete or simplify the key work underlying these patterns in advance and thus remove the need for exploration, implementation, and verification that the agent was meant to perform. To address this, we introduce TerminalHorizon, a fully automated data synthesis engine for long-horizon terminal agents. Starting from real-world professional work, it uses independent execution to identify key work omitted during construction and uses the resulting evidence to guide environment reconstruction. To extend these patterns across stages, it progressively expands tasks under the same goal, introducing new requirements that build on prior results and call for further exploration and improvement. Using this engine, we construct 1.5K task environments and perform supervised fine-tuning of Qwen3.5-27B on the resulting dataset to obtain TerminalHorizon-27B. We achieve substantial performance gains on all 12 benchmarks spanning coding-agent tasks, agentic tasks in real-world production settings, and general reasoning. Multiple RSI case studies, each with a 12-hour research budget, further demonstrate the model's ability to continue experimenting beyond an initial solution, improve its methods based on feedback, and retain effective improvements in its final deliverables.
4. 思维惯性：大语言模型被要求不思考时仍持续思考
- 英文标题：Thinking Inertia: LLMs Keep Thinking When Told Not To
- arXiv：2610.11765
💡 一句话解读：这篇论文研究的是：大语言模型即使被明确要求“别思考、直接给答案”，它还是忍不住在答案前写推理过程，就像人养成习惯后很难停下来。作者提出把回答拆成“答案前文字”和“最终答案”两部分，用三个指标衡量模型是否真的没思考，并在多种模型和题型上验证了这种“思维惯性”的存在。
大语言模型（LLM）越来越多地自带显式的“思考模式”，但其对立面“不思考”却鲜少受到关注。我们沿两个维度研究大语言模型的“不思考”行为。a. 如何衡量“不思考”？以往工作通常用禁用思考模式或缺少长推理链等替代指标来定义“不思考”，但这些替代指标并不可靠：被禁用的思考模式仍可能输出推理，而长推理链可能只是填充内容而非真正的推理。我们改为将每个回答规范化为“答案前推理片段”和“最终答案”，并在三个层面进行评估：(i) 空思考率（Empty-Thinking Rate），衡量严格只给答案的合规程度；(ii) 指令感知的问题—答案前相关性（Question-Pre-answer Relevance），衡量问题与答案前推理片段的相似度；(iii) 以LLM为评判者的显式推理率（Explicit Inference Rate），衡量可见的显式推理。这些指标共同区分了只输出答案、相关但非推理的文本以及显式推理三种情况。b. “不思考”在不同任务和模型间如何变化？我们在布尔题、选择题和开放题上，对六种提示干预和六个大语言模型进行了评估。我们发现，显式的“不思考”控制无法可靠地消除可见推理。模型反而表现出“思维惯性”（Thinking Inertia）：即使受到严格控制，显式推理依然存在，并且随着答案空间变得开放而更加普遍。在布尔题和选择题上准确率保持稳定，而开放题则揭示了“只给答案”的合规性与任务准确率之间的权衡。对相同问题在不同答案空间下进行改写表明，提供候选答案能让模型更容易只输出答案。这些发现确立了“不思考”是一项不可忽视的能力：不能仅凭模型设置或指令就假定模型会停止显式推理，它值得与推理能力一起被系统评估。
📄 English Abstract
Large Language Models (LLMs) increasingly ship with explicit "thinking modes", yet their counterpart, "no-thinking", has received far less attention. We study LLMs' no-thinking behavior along two axes. a. How to measure no-thinking? Prior work typically defines no-thinking through proxies such as a disabled thinking mode or the absence of long traces. These proxies are unreliable: disabled thinking modes may still emit reasoning, while long traces may contain filler rather than genuine inference. We instead normalize each response into a pre-answer trace and final answer, and evaluate it at three levels: (i) Empty-Thinking Rate for strict answer-only compliance; (ii) instruction-aware Question-Pre-answer Relevance for similarity between the question and pre-answer trace; and (iii) LLM-as-judge Explicit Inference Rate for visible explicit inference. Together, these metrics distinguish answer-only output, relevant but non-inferential text, and explicit inference. b. How does no-thinking vary across tasks and models? We evaluate six prompting interventions on six LLMs across Boolean, multiple-choice, and open-ended questions. We find that explicit no-think controls cannot reliably eliminate visible inference. Models instead exhibit "Thinking Inertia": explicit inference persists even under strict controls and becomes more prevalent as the answer space opens. Accuracy remains stable on Boolean and multiple-choice tasks, whereas open-ended tasks reveal a trade-off between answer-only compliance and task accuracy. Rewriting the same questions across answer spaces shows that supplying candidate answers makes answer-only responses easier to produce. These findings establish no-thinking as a non-trivial capability: stopping explicit reasoning cannot be assumed from model settings or instructions alone and deserves systematic evaluation alongside reasoning ability.
5. VIS-Ground：基于上下文锚定的视频交互式叙事
- 英文标题：VIS-Ground: Video Interactive Storytelling with Contextual Grounding
- arXiv：2610.09326
💡 一句话解读：观众在互动视频中提要求时，往往隐含了之前剧情和当前画面的依赖，但现有系统只盯着指令本身。本文提出VIS-Ground，把各种杂乱输入整理成可执行的约束，再通过规划、验证、修改和渲染来生成视频，在三种模型上平均提升10.3分。
视频交互式叙事让观众能主动引导视频的展开方式。然而，一旦允许观众在生成过程中介入，一个新挑战便会出现：观众的请求可能对锚定来源（grounding source）和当前已渲染的视频状态存在潜在依赖。这些依赖未必在任何单一输入中被明确表述，只有将来源、渲染历史与新的观众意图联合考虑时才会显现。现有的交互式视频生成系统主要强调遵循观众指令，而基于来源的视频生成方法则侧重于使生成内容与外部叙事或知识来源对齐。这使得一个根本问题尚未得到充分探索：在交互式续写过程中，生成模型应基于什么上下文进行锚定，以及如何将异构、非结构化的输入转化为这样的锚定上下文？在本工作中，我们将上下文锚定（contextual grounding）形式化为将异构输入上下文转化为可执行约束模型以用于视频生成的过程。为应对这一挑战，我们提出了VIS-Ground，它执行结构化上下文抽象（Structured Context Abstraction）以恢复锚定状态和跨上下文依赖，生成约束归纳（Generation Constraints Induction）以将相关依赖投影为候选特定的约束，以及约束视频生成（Constrained Video Generation）以通过规划、验证、修订和渲染来强制执行这些约束。在三种视频生成骨干模型上，VIS-Ground均持续取得最高的综合得分，相比每个骨干模型的最强基线平均绝对提升10.3分。详细分析进一步显示了在叙事锚定和知识锚定两方面的收益，并揭示了在依赖提取以及视频渲染过程中忠实实现方面仍存在的挑战。
📄 English Abstract
Video interactive storytelling enables viewers to actively steer how a video unfolds. However, once we allow viewers to intervene during generation, a new challenge arises: The viewer's request can have latent dependencies on both the grounding source and the current rendered video state. These dependencies may not be explicitly stated in any individual input, but emerge only when the source, rendered history, and new viewer intent are considered jointly. Existing interactive video generation systems primarily emphasize following viewer instructions, while source-grounded video generation methods focus on aligning generated content with an external narrative or knowledge source. This leaves a fundamental question underexplored: What context should a generation model ground on during interactive continuation, and how can heterogeneous, unstructured inputs be transformed into such grounding context? In this work, we formulate contextual grounding as the process of transforming heterogeneous input context into an executable constraint model for video generation. To address this challenge, we introduce VIS-Ground, which performs Structured Context Abstraction to recover grounded states and cross-context dependencies, Generation Constraints Induction to project relevant dependencies into candidate-specific constraints, and Constrained Video Generation to enforce these constraints through planning, verification, revision, and rendering. Across three video generation backbones, VIS-Ground consistently achieves the highest overall composite score, reaching an average absolute improvement of 10.3 points over the strongest per-backbone baselines. Detailed analysis further shows gains across both narrative and knowledge grounding, and reveals remaining challenges in dependency extraction, and faithful realization during video rendering.
6. 多流注意力残差
- 英文标题：Multi-Stream Attention Residuals
- arXiv：2610.13630
💡 一句话解读：这篇研究要解决的问题是：深层神经网络里，信息在层与层之间传递时容易被稀释或丢失。它提出的办法是让网络在“深度方向”上像多头注意力那样，从不同角度回看并挑选历史层的信息，再轻量地混合成当前层的输入，几乎不增加额外开销。
注意力残差（AttnRes）使网络能够沿深度方向进行依赖内容的检索，可视为沿深度轴的单流 softmax 注意力。受序列轴上多头注意力成功的启发，我们提出多流注意力残差（MSAR），将这一机制推广为多头深度注意力，在仅增加极小开销的情况下带来显著性能提升。MSAR 通过多尺度因果卷积从序列上下文中获取流的多样性，使不同头能够以不同的上下文视角检索历史表示，同时保持隐藏表示的完整性。检索到的各流通过轻量自适应混合进行组合，生成单流的层输入。通过保留单流来源并按需扩展，MSAR 无需持久的多流存储或通信即可支持多流计算。在 18B 和 28B MoE 规模的全部 12 个下游基准上，MSAR 均持续优于此前的残差变体，包括标准残差连接、流形约束超连接（mHC）和 AttnRes。缩放律实验显示，相较于标准残差基线，MSAR 实现了 1.58 倍的等损失计算加速，而 AttnRes 为 1.35 倍、mHC 为 1.30 倍。这些结果使 MSAR 成为大规模预训练中颇具前景的残差架构。
📄 English Abstract
Attention Residuals (AttnRes) enables content-dependent retrieval over network depth, which can be viewed as single-stream softmax attention along the depth axis. Inspired by the success of multi-head attention along the sequence axis, we introduce Multi-Stream Attention Residuals (MSAR), an extension of AttnRes that generalizes this mechanism to multi-head depth attention, yielding substantial performance gains with minimal additional overhead. MSAR draws stream diversity from sequence context through multi-scale causal convolutions, allowing different heads to retrieve historical representations through distinct contextual views while preserving the integrity of the hidden representation. The retrieved streams are combined through lightweight adaptive mixing to generate the single-stream layer input. By retaining single-stream sources and expanding them on demand, MSAR supports multi-stream computation without requiring persistent multi-stream storage or communication. Across all 12 downstream benchmarks at the 18B and 28B MoE scales, MSAR consistently outperforms prior residual variants, including standard residual connections, manifold-constrained Hyper-Connections (mHC), and AttnRes. Scaling-law experiments yield an iso-loss compute speedup of $1.58\times$ over the standard residual baseline, compared with $1.35\times$ for AttnRes and $1.30\times$ for mHC. These results make MSAR a promising residual architecture for large-scale pre-training.
7. 相关性不等于适用性：面向个人 GUI 智能体的经验激活
- 英文标题：Relevance Does Not Imply Applicability: Experience Activation for Personal GUI Agents
- arXiv：2609.33304
💡 一句话解读：个人GUI智能体用历史记录来理解模糊指令和预测习惯，但现有方法把任务相关的历史直接塞给模型，默认它对每一步都有用。研究发现，这些历史只在第一步有用，后面90%的步骤几乎没帮助。问题在于“相关”不等于“适用”：相关是任务层面的，适用是当前屏幕决定的。为此，作者提出ExpActivator，不训练模型，只在当前屏幕与历史状态匹配时才激活对应动作，并在时机合适时才主动建议。
个人图形用户界面（GUI）智能体依赖交互历史来从模糊指令中推断用户意图并预测重复性日常事务。现有方法检索任务相关的历史并将其附加到策略的上下文中，隐含地假设与任务相关的经验在该任务内的每个决策中仍然有用。我们发现这种帮助主要消耗在第一个决策上：检索到的历史显著改善了回合的开场步骤，但在其余90%的步骤中几乎没有持续收益，并且在判断是否有必要主动建议方面提供的指导很弱。一条相关的记录可能告诉智能体从哪里开始，但无法说明哪个过去动作适用于当前屏幕，或者某个日常事务现在是否该做。根本问题在于相关性并不意味着适用性：相关性在任务层面确定，而适用性取决于决策时的情境。因此，我们将个性化重新定义为经验激活，并引入ExpActivator，一个无需训练的框架，仅激活适用于当前情境的经验。在执行过程中，ExpActivator将每个新屏幕与冻结的GUI骨干网络潜在空间中的历史状态进行匹配，并提供相应的动作作为参考。在执行之前，仅当当前时间和场景提供足够支持时，它才激活重复性意图，否则弃权。在四个GUI骨干网络上，ExpActivator将轨迹内步骤成功率平均提高了28%，在每个骨干网络上实现了最佳的个性化执行，同时使用的历史令牌数约为五分之一，并且达到了最强主动基线的约2.3倍马修斯相关系数。经验在激活之处产生价值，而非在附加之处。
📄 English Abstract
Personal Graphical User Interface (GUI) agents rely on interaction history to infer what a user wants from ambiguous instructions and to anticipate recurring routines. Existing approaches retrieve task-relevant history and append it to the policy's context, implicitly assuming that experience relevant to a task remains useful for each decision within it. We find that this help is largely spent at the first decision: retrieved history strongly improves the opening step of an episode, yet provides little sustained benefit over the remaining 90% of steps, and offers weak guidance on whether a proactive suggestion is warranted. A relevant record may tell the agent where to begin, but not which past action applies to the current screen or whether a routine is due now. The underlying issue is that relevance does not imply applicability}: relevance is determined at the task level, whereas applicability depends on the situation at decision time. We therefore recast personalization as experience activation and introduce ExpActivator, a training-free framework that activates only the experience applicable to the current situation. During execution, ExpActivator matches each new screen to historical states in the frozen GUI backbone's latent space and supplies the corresponding action as a reference. Before execution, it activates a recurring intent only when the current time and scenario provide sufficient support, and otherwise abstains. Across four GUI backbones, ExpActivator improves within-trajectory step success by 28% on average, achieves the best personalized execution on every backbone while using about one-fifth as many history tokens, and reaches approximately 2.3$\times$ the Matthews correlation coefficient of the strongest proactive baseline. Experience pays where it is activated, not where it is appended.
8. CrossTimeEdit：面向历史街景生成的十年跨度跨视角数据集与奖励引导编辑
- 英文标题：CrossTimeEdit: A Decade-Spanning Cross-View Dataset and Reward-Guided Editing for Historical Street-View Generation
- arXiv：2609.36616
💡 一句话解读：历史街景照片覆盖不全，很多年代和地点缺失。本研究先造了一个横跨十年、11座城市、含4万多组配对的跨视角数据集，再把“生成过去街景”变成“编辑现有街景”：用近期照片锁定视角和不变部分，用卫星影像差异提示哪里变了，并通过奖励引导训练模型，让生成的旧街景既符合历史变化又真实自然。
历史街景影像记录了城市演变，但覆盖不均导致历史记录存在大量空白。要生成合理的往昔街景，需在保留未变场景内容的同时还原已变化的结构。我们构建了 VIGOR-his，一个跨越十年的跨视角数据集，包含三大洲 11 座城市的 43,653 个地点级四元组；其自动化流程完成空间配对、一致性筛选、变化分类，以及基于卫星影像的变化描述与局部编辑指令的生成与验证。基于 VIGOR-his，我们提出 CrossTimeEdit，将历史街景生成重构为编辑任务，用近期街景约束视角与未变外观，用时序卫星差异作为变化证据。我们从 FLUX.2 [Klein] 4B 出发，通过监督微调（SFT）和在线强化学习（RL）训练 CrossTimeEdit。我们设计了三项街景编辑准则：指令对齐（IA）、背景保持（BP）与质量及物理合理性（QP），既作为 RL 奖励维度也作为评估指标。我们采用面向流匹配模型的组相对策略优化（Flow-GRPO）与组奖励解耦归一化策略优化（GDPO）来优化这一多奖励目标，后者在聚合前对每个奖励维度分别归一化。CrossTimeEdit 在三项编辑准则上的整体表现较预训练基线提升 17.12%，并在场景一致性、视觉真实感和感知质量上优于跨视角生成模型。实现代码、数据集与模型权重见 https://luhanwen67.github.io/CrossTimeEdit-release/。
📄 English Abstract
Historical street-view imagery records urban evolution, but uneven coverage leaves substantial gaps in historical records. Generating plausible past appearances requires restoring changed structures while preserving persistent scene content. We construct VIGOR-his, a decade-spanning cross-view dataset containing 43,653 location-level quadruplets across 11 cities on three continents. Its automated pipeline performs spatial pairing, consistency screening, change classification, and the generation and validation of satellite-based change descriptions and local editing instructions. Based on VIGOR-his, we propose CrossTimeEdit, a model that reformulates historical street-view generation as editing, using recent street views to constrain viewpoint and unchanged appearance and temporal satellite differences as change evidence. Starting from FLUX.2 [Klein] 4B, we train CrossTimeEdit through supervised fine-tuning (SFT) followed by online reinforcement learning (RL). We design three street-view editing criteria, namely Instruction Alignment (IA), Background Preservation (BP), and Quality and Physical Plausibility (QP), as both RL reward dimensions and evaluation metrics. We optimize this multi-reward objective using Within Group Relative Policy Optimization for flow-matching models (Flow-GRPO) with Group reward-Decoupled Normalization Policy Optimization (GDPO), which normalizes each reward dimension before aggregation. CrossTimeEdit improves overall performance across the three editing criteria by 17.12% over the pretrained baseline and outperforms cross-view generation models in scene consistency, visual realism, and perceptual quality. The implementation code, dataset, and model weights are available at https://luhanwen67.github.io/CrossTimeEdit-release/.
9. GroundAnything：以闪电速度协调并行解码与精确视觉定位
- 英文标题：GroundAnything: Reconciling Parallel Decoding with Precise Visual Grounding at Flash Speed
- arXiv：2609.39600
💡 一句话解读：传统目标定位模型像逐字写句子一样从左到右依次输出位置，速度慢且顺序不自然。这篇论文把定位看作从图像中提取证据，改用扩散模型并行生成多个候选框并逐步修正，还通过特殊训练让模型又快又准。
自回归（AR）定位模型将空间预测序列化，引入了顺序延迟并对输出词元施加了因果顺序。我们将定位视为视觉证据提取：物体、位置和空间关系由图像和查询共同约束，但它们之间的依赖关系并不意味着存在固有的从左到右生成顺序。这一区别使得双向扩散成为自然选择，允许空间假设并行出现并通过迭代去噪共同细化。我们提出了 GroundAnything，一个 4B 参数的定位基础模型，通过分块去噪协调快速并行解码与精确定位。训练结合了来自公共数据集和专用数据引擎的定位预训练、带有联合 AR 和扩散目标的直接 AR 到扩散转换、监督微调以及基于 GRPO 的强化后训练。在 30 个定位基准上，我们的自回归变体 GroundAnything-VLM 在同等规模模型中取得了新的总体最优结果，达到 72.42%，与 GPT-6 Astra（71.35%）保持竞争力。借助熵引导解码，GroundAnything 还超越了该规模下的先前最优，平均达到 61.75%，而基于快速 MTP 的 LocateAnything 模型为 53.32%。我们进一步探索了解码策略，表明可选的自推测模式相比 AR 对应模型实现了 4.51 倍加速，COCO F1mIoU 仅下降 0.74 个百分点。基础设施实验表明，渐进式推理优化将并行解码转化为实际加速。这些支持了延迟敏感的真实世界系统中的高效视觉定位。
📄 English Abstract
Autoregressive (AR) grounding models serialize spatial predictions, introducing sequential latency and imposing a causal order on output tokens. We view grounding as visual evidence extraction: objects, locations, and spatial relations are jointly constrained by the image and query, yet their dependencies do not imply an intrinsic left-to-right generation order. This distinction makes bidirectional diffusion a natural fit, allowing spatial hypotheses to emerge in parallel and be jointly refined through iterative denoising. We introduce GroundAnything, a 4B-parameter grounding foundation model that reconciles fast parallel decoding with precise localization through blockwise denoising. Training combines grounding pretraining from public datasets and dedicated data engines, direct AR-to-diffusion conversion with joint AR and diffusion objectives, supervised fine-tuning, and GRPO-based reinforcement post-training. Across 30 grounding benchmarks, our autoregressive variant, GroundAnything-VLM, establishes a new overall state of the art among similarly sized models at 72.42%, remaining competitive with GPT-6 Astra (71.35%). With entropy-guided decoding, GroundAnything also surpasses the prior state of the art at this scale, averaging 61.75% versus 53.32% for the fast MTP-based LocateAnything model. We further explore decoding strategies, showing that an optional self-speculative mode achieves a $4.51\times$ speedup over the AR counterpart with a 0.74 percentage-point drop in COCO F1mIoU. Infrastructure experiments show that progressive inference optimizations translate parallel decoding into practical speedups. These support efficient visual grounding in latency-sensitive real-world systems.
10. SpatialASM：内化主动心智化以实现空间推理
- 英文标题：SpatialASM: Internalizing Active Mentalization for Spatial Reasoning
- arXiv：2610.13832
💡 一句话解读：当前的大型视觉语言模型在空间推理时不会像人一样主动切换视角、在脑中重建场景。这篇论文提出 SpatialASM 框架，通过训练时的空间模拟，让模型学会自己规划并想象新视角，从而提升空间推理能力。
在解决复杂空间推理任务时，普通人会自然地从与任务相关的视角重新审视场景，并在脑中重建这些视角下的空间布局。但大型视觉语言模型（LVLMs）能否以同样方式主动思考？我们将这种能力称为主动空间心智化（Active Spatial Mentalization, ASM），它是视觉智能的内在组成部分，而当前 LVLMs 在这方面仍远未发展。为弥补这一差距，我们提出 SpatialASM，一个可扩展的框架，通过训练时的空间模拟将 ASM 内化到 LVLMs 中。SpatialASM 采用一个异中心新视角生成器（allocentric novel-view generator）将假设视角锚定在视觉证据上，并产生两种互补的监督形式：i）显式 ASM 轨迹，教 LVLMs 在空间推理中何时以及如何自适应地调用视角规划与想象；ii）自动构建的视角采择任务，进一步增强以视角为条件的场景想象。在涵盖从相对感知导向的空间理解到需要多图像、跨视角和视角变换推理的更具挑战性任务的七个基准上，SpatialASM 在四种 LVLM 主干上提升了空间推理性能，同时保留了它们原有的通用视觉能力。实验结果表明，我们的方法提高了七个空间基准上的平均准确率。并且，ASM 轨迹的合成提供了一条可扩展的途径，以缓解空间思维链（Chain-of-Thought, CoT）监督的稀缺性，从而激发主动视角探索和心理场景模拟。
📄 English Abstract
When solving complex spatial reasoning tasks, ordinary humans naturally revisit the scenes from task-relevant viewpoints and mentally reconstruct the spatial layouts from those viewpoints. But can Large Vision-Language Models (LVLMs) proactively think in the same way? We term this ability Active Spatial Mentalization (ASM), an intrinsic component of visual intelligence that remains substantially underdeveloped in current LVLMs. To bridge this gap, we introduce SpatialASM, a scalable framework to internalize ASM into LVLMs through training-time spatial simulation. SpatialASM employs an allocentric novel-view generator to ground hypothetical viewpoints in visual evidence and produces two complementary forms of supervision: i) explicit ASM trajectories that teach LVLMs when and how to adaptively invoke view planning and imagination during spatial reasoning, and ii) automatically-constructed perspective-taking tasks that further strengthens viewpoint-conditioned scene imagination. On seven benchmarks spanning relatively perception-oriented spatial understanding to more challenging tasks requiring multi-image, cross-view and viewpoint-transformation reasoning, SpatialASM yields performance gain on LVLMs of four LVLM backbones in spatial reasoning while retaining their original general vision capabilities. The experimental results demonstrate that our approach improves average accuracy across the seven spatial benchmarks. And the synthesis of ASM trajectories provides a scalable pathway to alleviate the scarcity of spatial Chain-of-Thought (CoT) supervision for eliciting active viewpoint exploration and mental scene simulation.
11. 系统化多智能体视觉与语言导航：形式化、基准与方法
- 英文标题：Systematic Multi-Agent Vision-and-Language Navigation: Formulation, Benchmark, and Method
- arXiv：2609.35965
💡 一句话解读：现有视觉语言导航只让单个机器人听一条指令行动，但现实中很多任务需要多个机器人合作才能完成。这篇论文首次把多机器人协作导航形式化为带依赖和资源约束的协调问题，并构建了包含11724个任务、最多四个机器人组队的基准数据集MAVLN，还提出一个能调度、共享记忆和避碰的基线系统TRISS。
视觉与语言导航（VLN）此前主要关注单个智能体遵循单条指令，但许多现实应用需要机器人团队来完成超出任何单个智能体能力的任务。我们提出系统化多智能体视觉与语言导航，据我们所知，这是首次将多智能体VLN系统化地形式化为一个带约束的协调问题：每个任务由携带依赖与资源约束（存在锁与持有链）的子任务组成。一个经过验证的四阶段构建流程将该任务实例化为MAVLN，包含145个场景中的11724个回合，团队规模最多四个智能体，涵盖三种指令模式，并配有定制的约束感知指标。我们进一步提出TRISS，一个面向协调的导航系统，它耦合了基于大语言模型的子任务调度器、将每个智能体的探索转化为团队知识的共享拓扑记忆，以及将同时意图实现为无碰撞路径的冲突感知执行机制。大量实验确立了TRISS作为一个全面的基线，并揭示了在调度、规划与执行方面仍有很大改进空间，凸显了在MAVLN任务约束下进行协调的挑战。项目页面：https://xyz9911.github.io/mavln。
📄 English Abstract
Vision-and-Language Navigation (VLN) has largely focused on a single agent following a single instruction, yet many real-world applications require teams of robots to tackle tasks beyond the capabilities of any individual agent. We present Systematic Multi-Agent Vision-and-Language Navigation, providing, to our knowledge, the first systematic formalization of multi-agent VLN as a constrained coordination problem: each mission consists of subtasks carrying dependency and resource constraints (presence locks and holding chains). A verified four-stage crafting pipeline instantiates the task as MAVLN, comprising 11,724 episodes across 145 scenes with teams of up to four agents under three instruction regimes, accompanied by tailored constraint-aware metrics. We further present TRISS, a coordination-ready navigation system coupling an LLM-based subtask scheduler, a shared topological memory that turns each agent's exploration into team knowledge, and a conflict-aware execution mechanism that realizes simultaneous intentions as collision-free routes. Extensive experiments establish TRISS as a comprehensive baseline and reveal substantial room for improvement across scheduling, planning, and execution, highlighting the challenges of coordinating under MAVLN task constraints. Project page: https://xyz9911.github.io/mavln.
12. T3lescope：基于图像的任意分辨率高保真生成式表面重建
- 英文标题：T3lescope: Arbitrary-Resolution High-Fidelity Generative Surface Reconstruction from Images
- arXiv：2610.03308
💡 一句话解读：从多视角照片重建3D场景时，现有方法要么需要逐场景慢慢优化、遇到稀疏视角或反光透明表面就失效，要么只能固定分辨率处理小块区域，导致大场景难以保持一致。T3lescope用一个固定分辨率的生成模型，在推理时从粗到细逐级细化：先定大布局，再在越来越小的空间单元里补充细节，权重共享、层级灵活，从而高保真重建从单个物体到城市级的大场景。
我们无需逐场景优化，即可从带位姿的多视角图像中重建高保真三维场景网格，覆盖从单个物体到大型户外场景的多种尺度。逐场景优化方法在观测稀疏或表面光滑、透明时，缺乏所需的已学习三维先验。现有生成式方法利用此类先验补全稀疏观测区域的几何，但通常在有限空间范围内以固定分辨率运行，在空间覆盖与细节之间取舍。因此，重建大型场景往往需要将其划分为独立处理的重叠局部区域，难以保持全局几何一致性。为解决这些问题，我们提出 T3lescope，它在推理时以由粗到细的级联方式，将单个固定分辨率生成器应用于不同场景尺度。粗层级确立场景布局，更细层级在逐步变细的空间单元内对继承自粗层级的几何进行扰动与去噪，以恢复表面细节。该模型在多个尺度上以单个单元训练，并在所有层级间共享权重，因此训练时不固定层级结构，层级数量、单元尺度与单元位置均在推理时确定。在室内、户外和城市级场景上，T3lescope 优于前馈与生成式基线方法，达到或超过逐场景优化，并能恢复精细结构以及光滑和透明表面。这些结果表明，我们的方法可泛化到多样场景、视角数量和图像分辨率。项目主页：https://pfnet-research.github.io/t3lescope/
📄 English Abstract
We reconstruct high-fidelity 3D scene meshes from posed multi-view images without per-scene optimization, across scales ranging from single objects to large outdoor scenes. Per-scene optimization methods lack the learned 3D prior needed when observations are sparse or surfaces are glossy or transparent. Existing generative methods leverage such priors to complete geometry in sparsely observed regions, but typically operate at a fixed resolution over a limited spatial extent, trading spatial coverage against detail. Reconstructing a large scene therefore often requires partitioning it into independently processed overlapping local regions, making it difficult to maintain global geometric consistency. To address these issues, we propose T3lescope, which applies a single fixed-resolution generator across scene scales in an inference-time coarse-to-fine cascade. A coarse level establishes the scene layout, and finer levels perturb and denoise geometry inherited from the coarser level within progressively finer spatial cells to recover surface detail. The model is trained on individual cells at multiple scales and shares its weights across all levels, so no hierarchy is fixed during training, and the number of levels, cell scales, and cell locations are determined at inference time. On indoor, outdoor, and city-scale scenes, T3lescope outperforms feed-forward and generative baselines, matches or surpasses per-scene optimization, and recovers fine structures as well as glossy and transparent surfaces. These results show that our method generalizes across diverse scenes, view counts, and image resolutions. Project page: https://pfnet-research.github.io/t3lescope/
13. 以更少训练学习更多：面向高效推理后训练的不确定性引导渐进收缩
- 英文标题：Learning More by Training Less: Uncertainty-Guided Progressive Contraction for Efficient Reasoning Post-Training
- arXiv：2610.12933
💡 一句话解读：这篇研究想解决大模型推理训练中数据越多、成本越高但效果提升不成比例的问题。它提出一种方法：先围绕推理中间步骤生成多种可能路径，找出最不确定、最值得学的部分，把训练数据大幅压缩；再根据学生模型的学习状态动态挑选样本，从而用极少的数据和计算量达到甚至超过全量训练的效果。
大语言模型在复杂推理方面已取得显著进展。然而，用越来越大的语料库扩展推理后训练，往往会大幅增加训练成本，却无法带来成比例的性能提升。为此，本研究提出一个不确定性引导的渐进收缩框架，将监督和优化集中在最有学习价值的推理轨迹上。具体而言，我们不再将教师生成的思维链（CoT）视为单一实现路径，而是围绕中间推理状态构建蒙特卡洛推理树，以暴露过程级不确定性，从而构建带有不确定性画像的推理监督。该不确定性首先将原始语料收缩为一个紧凑的教师知情池；随后，第二个感知学习者的收缩在监督微调（SFT）期间进行在线路由，根据学生不断变化的学习状态集中优化。两个阶段共同渐进地收缩监督与优化，大幅降低学生训练预算。在多种推理基准和LLM系列上的实验表明，我们的方法仅使用全数据SFT优化预算的0.22%，训练曝光量减少450倍以上，同时保持或提升推理性能。值得注意的是，该框架在学生模型系列异构、教师为黑盒以及弱到强迁移等场景下均表现良好，这表明它不仅能加速训练，还能增强推理能力。
📄 English Abstract
Large language models have achieved remarkable advances in complex reasoning. However, scaling reasoning post-training with ever-larger corpora can substantially increase training cost without proportional performance gains. In this study, we therefore ask: how far can reasoning supervision be compressed without sacrificing capability? To this end, we propose an uncertainty-guided progressive contraction framework that concentrates supervision and optimization on the reasoning trajectories most valuable for learning. Specifically, rather than treating a teacher-generated chain-of-thought (CoT) as a single realized path, we construct uncertainty-profiled reasoning supervision by building Monte Carlo reasoning trees around intermediate reasoning states to expose process-level uncertainty. This uncertainty first contracts the original corpus into a compact teacher-informed pool. A second, learner-aware contraction then performs online routing during SFT, concentrating optimization according to the student's evolving learning state. Together, the two stages progressively contract both supervision and optimization, substantially reducing the student training budget. Experiments across diverse reasoning benchmarks and LLM families show that our method uses as little as 0.22% of the optimization budget of full-data SFT, over 450x less training exposure, while maintaining or improving reasoning performance. Notably, our framework generalizes well across heterogeneous student families, black-box teachers, and weak-to-strong transfer. This indicates that it is not only a way to accelerate training but can also enhance reasoning capability.
14. ARCS：通过结构化消歧迈向精准的文本到SQL
- 英文标题：ARCS: Towards Precise Text-to-SQL via Structured Disambiguation
- arXiv：2610.09396
💡 一句话解读：用户提问模糊时，文本到SQL系统容易出错。现有方法靠对话澄清，效率低。本文提出结构化消歧，用受限交互代替自由对话，并构建首个真实数据库模糊性基准ARCS，实验显示现有模型准确率很低。
随着文本到SQL（text-to-SQL）系统从演示走向实际部署，用户问题中的模糊性成为主要错误来源。这类模糊性往往细微、与领域或数据相关，并可能悄无声息地使系统输出偏离用户真实意图。传统上通过对话澄清来解决模糊性，但这种方式效率低、认知负担重，且与实际用户工作流程不匹配。我们提出结构化消歧（structured disambiguation），一种通过显式、受限交互而非自由对话来解决模糊性的新范式。我们构建了ARCS（面向SQL的模糊性消解语料库），这是首个针对真实数据库、包含自然出现且不受约束的模糊性的文本到SQL基准，并完整标注了所有有效模糊点、解释和SQL查询。实验结果表明，存在模糊性时文本到SQL仍具挑战性：gpt-6-sol仅达到51%的端到端执行准确率，且没有开源模型超过27%。
📄 English Abstract
As text-to-SQL systems move beyond demonstrations toward real-world deployment, ambiguity in user questions becomes a primary source of errors. Such ambiguities are often subtle, domain- or data-specific, and can silently cause system outputs to deviate from the user's true intent. Ambiguity is traditionally addressed through conversational clarification, which is often inefficient, cognitively demanding, and poorly aligned with real-world user workflows. We propose structured disambiguation, a new paradigm in which ambiguity is resolved through explicit, constrained interactions rather than free-form dialogue. We construct ARCS (Ambiguity Resolution Corpus for SQL), the first text-to-SQL benchmark featuring naturally occurring, unconstrained ambiguities over real-world databases, with complete annotations of all valid ambiguity points, interpretations, and SQL queries. Experimental results show that text-to-SQL remains challenging in the presence of ambiguity: gpt-6-sol achieves only 51% end-to-end execution accuracy, and no open-source model exceeds 27%.
15. SPLATIFY：复现、发现、创新！从论文与想法到可训练的3DGS代码
- 英文标题：SPLATIFY: Reproduce, Discover, Innovate! From Papers and Ideas to Trainable 3DGS Code
- arXiv：2610.09116
💡 一句话解读：这篇论文要解决的是：3D高斯泼溅（3DGS）领域论文太多，想在此基础上做研究就得先花几周把论文重新写成能跑的代码。SPLATIFY 用多个 AI 智能体协作，自动把论文变成可训练的代码，还能自己找弱点改进效果、跨学科借鉴物理知识发明新方法，把复现时间从几周缩短到几分钟。
3D高斯泼溅（3DGS）研究的快速增长，使得人们在基于已有工作继续推进之前，必须投入大量精力重新实现论文。我们提出 SPLATIFY，一个多智能体框架，可将 3DGS 论文转换为基于 gsplat 的可训练实现，而通用的论文转代码方法和前沿模型在此任务上均告失败。SPLATIFY 通过五项创新实现这一目标：(1) 面向 gsplat 的上下文无关文法，建立在带有损失、致密化、渲染与优化扩展点的模块化方法模板之上，通过构造约束生成代码，使其天然满足 gsplat 的架构不变式。(2) 用于忠实复现的架构要素：分叉感知的引用恢复，以函数级粒度检索组件级代码；按拓扑依赖顺序进行的思维图（Graph-of-Thought）合成；从 20 多个已验证实现中进行的 RAG 引导的上下文示例选择；以及结合 PSNR 引导重生成、高斯级结构检查与 VLM 驱动补丁的视觉反馈。(3) 知识驱动的组合式改进，可自主发现弱点，并组合互补的正则化器、损失与致密化策略，以超越原始结果。(4) 跨学科方法发现，智能体从 3DGS 文献之外检索物理先验，并将其与渲染知识组合，为此前未被处理的场景类型生成方法。(5) SPLATIFY-Bench，一个覆盖 30 篇多样化 3DGS 论文的评测框架。在无公开代码的论文上，SPLATIFY 达到专家实现水平，同时将开发时间从数周缩短至数分钟，并通过组合式发现将 PSNR 进一步提升最多 2.4 dB。我们还展示了由 SPLATIFY 完全合成的、用于体积星云渲染及其他科学领域的新方法。
📄 English Abstract
The rapid growth of 3D Gaussian Splatting (3DGS) research demands significant effort to reimplement papers before building on them. We introduce SPLATIFY, a multi-agent framework that converts 3DGS papers into trainable gsplat-based implementations, where generic paper-to-code methods and frontier models fail. SPLATIFY achieves this through five innovations: (1) A context-free grammar for gsplat over a modular method template with extension points for losses, densification, rendering, and optimization, constraining synthesis so generated code satisfies gsplat's architectural invariants by construction. (2) Architectural elements for faithful reproduction: fork-aware citation recovery retrieving component-level code at function-level granularity, Graph-of-Thought synthesis in topological dependency order, RAG-guided in-context example selection from over 20 verified implementations, and visual feedback combining PSNR-guided regeneration, Gaussian-level structural checks, and VLM-driven patching. (3) Knowledge-driven compositional improvement that autonomously finds weaknesses and composes complementary regularizers, losses, and densification strategies to improve upon original results. (4) Interdisciplinary method discovery where agents retrieve physical priors from outside the 3DGS literature and compose them with rendering knowledge to produce methods for previously unaddressed scene types. (5) SPLATIFY-Bench, an evaluation framework across 30 diverse 3DGS papers. On papers without public code, SPLATIFY matches expert implementations while reducing development time from weeks to minutes, and through compositional discovery further improves PSNR by up to 2.4 dB. We additionally demonstrate novel methods for volumetric nebula rendering and other scientific domains, synthesized entirely by SPLATIFY.