为解决这一问题，复旦大学、南洋理工大学的研究者提出一种基于视觉 Grounding 的多轮强化学习方法 MGPO，使 LMM 能在多轮交互中根据问题，自动预测关键 区域坐标，裁剪子图像并整合历史上下文，最终实现高分辨率图像的精准推理。相比监督微调（SFT）需要昂贵的 Grounding 标注作为监督，MGPO 证明了在强化 学习（RL）范式中，即使没有 Grounding 标注，模型也能从 「最终答案是否正确」的反馈中，涌现出鲁棒的视觉 Grounding 能力。 MGPO 的核心创新点包括： 1） 自上而下的可解释视觉推理： 赋予了 LMMs 针对高分辨率场景的 「自上而下、问题驱动」 视觉搜索机制，提供可解释的视觉 Grounding 输出； 2） 突破最大像素限制： 即使因视觉 Token 数受限导致高分辨率图像缩放后模糊，模型仍能准确识别相关区域坐标，从原始高分辨率图像中裁剪 出清晰子图像用于后续分析； 3） 无需额外 Grounding 标注： 可直接在标准 VQA 数据集上进行 RL 训练，仅基于答案监督就能让模型涌现出鲁棒的视觉 Grounding 能力。 标题： High-Resolution ...

点击下方 卡片 ，关注" 具身智能 之心 "公众号 一、你的3D Grounding 模型真的work吗？ 在具身智能系统中，智能体需要依靠第一视角的3D感知算法来理解周边环境。作为其中的核心任务之一，Embodied 3D Grounding是指根据ego-centric的RGB- D图像序列以及语言描述在三维空间中定位目标对象，要求模型能够融合语言与三维视觉信息，准确识别出语句中所指代的物体。当前主流方法多采用两阶段策 略，即先利用检测模型提取三维区域特征，再进行语言引导的grounding微调。这自然引出一个疑问： 第二阶段这种针对 Grounding 的微调，其效果究竟如何，它 真的work吗？ 令人颇感意外的是，实证结果显示，即便是当前最先进的Grounding模型，其实际表现也远未达到预期。相反，那些完全未接受语言监督、仅依赖目标类别进行筛 选的检测模型，在Grounding任务的评估中竟取得了更优的结果。具体而言，考虑到任务中的语言指令为模板生成，本文通过规则解析提取出目标物体的类别标 签，之后使用该类别从检测模型中筛选对应预测框，直接作为Grounding的输出。理论上，这种做法缺乏语言理解过 ...

Core Idea - Factory believes software development is transitioning to agent-driven from human-driven [1] - To achieve significant productivity gains (5-20x), a shift from collaborating with AI to delegating tasks entirely to AI is needed [3] - Factory is building a platform for managing and scaling AI agents, integrating various engineering systems [3][4][5] Agentic System Characteristics - Agentic systems require planning to decide future actions [11] - Decision-making is crucial for agents to make calls based on the existing state [13][14] - Environmental grounding is necessary for agents to interact with and adapt to the external environment [14] Human-AI Collaboration - Humans will remain in software development, focusing on the outer loop (reasoning, requirements) [15][16] - Agents will handle the inner loop (coding, testing, code review) [17] - AI UX should blend delegation with control for situations where agents cannot complete tasks [17] Agent Reliability - Clear planning and boundaries are essential for reliable agents [32] - Subtask decomposition, model predictive control, and explicit plan templating can improve planning [19][20] - Control over the tools agents use is the most important differentiator in agent reliability [28] Environmental Interaction - New AI computer interfaces are needed for agents to interact with the world [28] - Processing information from the environment is crucial for complex systems [29][30] - Agents need to ground themselves in the environment to perform full software development work [32] Call to Action - Factory encourages teams not delegating at least 50% of engineering tasks to AI agents to engage with them [34]

Core Insights - Tesla Grounding is a wearable energy harmonization device utilizing scalar wave technology inspired by Nikola Tesla, aimed at individuals experiencing energetic imbalances due to EMF exposure and modern lifestyle stressors [2][10][12] - The product offers a non-invasive, portable solution for grounding without the need for wires or batteries, making it suitable for various environments [5][13][17] Group 1: Product Overview - Tesla Grounding operates on the principle of bioenergetic coherence, designed to harmonize the body's natural electrical state with environmental frequencies [14][20] - The device is lightweight, waterproof, and requires no maintenance, allowing for easy integration into daily routines [17][65] - It is positioned as a modern alternative to traditional grounding methods, which often require physical contact with the Earth [29][63] Group 2: Target Audience - Ideal users include professionals in high-EMF environments, frequent travelers, energy-sensitive individuals, holistic wellness practitioners, students, and seniors seeking gentle wellness support [70][72][76] - The product is particularly beneficial for those experiencing symptoms of energetic fatigue, such as tension and mental fog, due to prolonged exposure to electronic devices [26][69] Group 3: User Experiences - User testimonials highlight benefits such as improved energetic alignment, stress relief, and enhanced vibrational coherence, although these experiences are anecdotal [48][49][56] - Many users report feeling more grounded and emotionally centered, especially in environments with high electromagnetic exposure [54][56] Group 4: Pricing and Warranty - Tesla Grounding is available in various pricing packages, with a total price ranging from $399 for a single unit to $999 for a family pack [80][81] - The product comes with a 30-day money-back guarantee, allowing customers to evaluate its compatibility with their wellness needs risk-free [84][89]