2026.06.19 | RATs让机器人自主玩耍学技能；Moebius用0.2B参数实现10B级修复性能

【赞助商】OpenClaw快报每天五分钟，听听 OpenClaw 快报，带你了解最新动态和业内讨论传送门 https://www.xiaoyuzhoufm.com/podcast/6a1732a2dffa135d0ab5ef43【目录】本期的 15 篇论文如下：[00:33] 🤖 Playful Agentic Robot Learning（趣味自主型机器人学习）[01:22] 🎨 Moebius: 0.2B Lightweight Image Inpainting Framework with 10B-Level Performance（Moebius：0.2B轻量级图像修复框架，实现10B级性能）[02:10] 🧠 S-Agent: Spatial Tool-Use Elicits Reasoning for Spatial Intelligence（S-Agent：空间工具使用激发空间智能推理）[03:10] 📊 Beyond Static Leaderboards: Predictive Validity for the Evaluation of LLM Agents（超越静态排行榜：LLM智能体评估的预测有效性）[04:05] 🎨 FreeStyle: Free Control of Style-Content Dual-Reference Generation from Community LoRA Mining（FreeStyle：基于社区LoRA挖掘的自由风格-内容双参考生成控制）[05:06] 🪄 JanusMesh: Fast and Zero-Shot 3D Visual Illusion Generation via Cross-Space Denoising（JanusMesh：通过跨空间去噪实现快速且零样本的3D视觉错觉生成）[05:58] 🤖 ENPIRE: Agentic Robot Policy Self-Improvement in the Real World（ENPIRE：实体世界中智能体机器人策略的自我改进）[06:57] 👁 Thinking with Visual Grounding（视觉锚定思考）[07:41] 🔍 Understanding the Behaviors of Environment-aware Information Retrieval（理解环境感知信息检索的行为）[08:37] 🤖 FAPO: Fully Autonomous Prompt Optimization of Multi-Step LLM Pipelines（FAPO：多步骤大语言模型管道的全自主提示优化）[09:28] 🧊 Adaptive Volumetric Mechanical Property Fields Invariant to Resolution（自适应体积力学属性场，分辨率不变性）[10:23] 📸 DF3DV-1K: A Large-Scale Dataset and Benchmark for Distractor-Free Novel View Synthesis（DF3DV-1K：用于无干扰新视角合成的大规模数据集与基准）[11:16] 🌍 Holo-World: Unified Camera, Object and Weather Control for Video World Model（全息世界：面向视频世界模型的统一相机、物体与天气控制）[12:12] 🎨 ImageWAM: Do World Action Models Really Need Video Generation, or Just Image Editing?（ImageWAM：世界动作模型真的需要视频生成，还是只需图像编辑？）[13:07] 🎯 Selective Synergistic Learning for Video Object-Centric Learning（面向视频对象中心学习的选择性协同学习）【关注我们】您还可以在以下平台找到我们，获得播客内容以外更多信息小红书: AI速递在小宇宙查看该单集文稿