2026.11.20 | 诊断机器人思行断点;声音画面联合理解

2026.11.20 | 诊断机器人思行断点;声音画面联合理解

10分钟 ·
播放数22
·
评论数0

⏱️ Highlights
00:28 🤖 Mind2Act World: A Diagnostic Benchmark for Embodied Mind, Physical Action, and Mind-Action Composition in Robotic Manipulation(Mind2Act World:机器人操作中具身心智、物理动作与心智—动作组合的诊断基准)
01:03 🎬 WeMM-Video: Pushing the Frontiers of Open-Source Omni-Modal Video Understanding(WeMM-Video:推进开源全模态视频理解的前沿)
01:37 🔎 WeVisForensics: Self-Evolving Agent for Omni AI-Generated Video Detection(WeVisForensics:面向全模态AI生成视频检测的自进化智能体)
02:14 🎨 Coda-DaVinci: Reimagining Video Generation with Code2Video(Coda-DaVinci:用Code2Video重新构想视频生成)
02:46 🧩 The Mask Is Not the Object: Volumetric Supervision for 3D Gaussian Segmentation(掩码不等于物体:面向三维高斯分割的体积监督)
03:19 👁️ EgoCaptioner: Towards Comprehensive and Fine-Grained Egocentric Video Captioning(EgoCaptioner:迈向全面、细粒度的第一视角视频描述)
03:55 🖱️ TuringSpace: Grounding Spatial Intelligence in Action with Interactive CAPTCHAs(TuringSpace:用交互式验证码将空间智能落实到行动中)
04:30 🖼️ HiUni: Harmonizing Understanding and Generation in Shared Visual Space(HiUni:在共享视觉空间中协调理解与生成)
05:06 🧭 ZeroVLN: Learning to Navigate without Supervised Fine-Tuning of the VLM Brain(ZeroVLN:无需对VLM大脑进行监督微调的导航学习)
05:42 🎲 Posterior Matching Distillation(后验匹配蒸馏)
06:18 📏 Depth Pro 2: Sharp and Temporally Consistent Metric Monocular Depth(Depth Pro 2:清晰且时间一致的单目公制深度)
06:52 🐾 WAM-Policy: World Action Models with On-Policy Distillation for Quadrupedal Loco-Manipulation(WAM-Policy:面向四足移动操作、结合在策略蒸馏的世界动作模型)
07:26 🧠 Probing the "Conscious" Mind: Selective Access and Global Workspace in Vision-Language Models(探查“有意识”的心智:视觉语言模型中的选择性访问与全局工作空间)
07:59 🌍 WM^MLLM: Turning MLLMs into World Models(WM^MLLM:将多模态大语言模型转变为世界模型)
08:37 ⚖️ AltRA-Test: An Alternative Annotator Test for LLM Judges Without Shared Annotators(AltRA-Test:无需共享标注者的LLM裁判替代标注者检验)

【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