宇树G1完成全球首例机器人活体手术登Nature | Unitree G1 Performs First Robotic Live Surgery, Featured in Nature
一台造价不到10万的人形机器人，在远程操控下完成了猪腹腔镜胆囊切除——这是通用人形机器人首次在活体手术场景留下印记。
宇树G1登上Nature：全球首例通用人形机器人活体微创手术
7月14日，国际顶刊《Nature》刊发重磅论文：基于国产宇树G1通用人形机器人，完成了全球首例通用人形机器人远程操控活体猪腹腔镜胆囊切除微创手术。
这项成果填补了通用人形机器人在精密医疗场景的空白。宇树G1作为一款售价不到10万元人民币的通用人形机器人，通过远程操控完成了需要高精度操作的微创手术流程。
智源Orca世界模型：12.5万小时无标注视频训练
同日，智源研究院发布悟界·RoboBrain Orca通用世界基座模型。该模型依托12.5万小时海量无标注真实视频完成预训练，不依赖机器人动作标签，下游机器人任务综合性能追平各类专用具身系统。
这意味着通用世界模型在机器人领域的应用迈出了关键一步——不再需要为每个具体任务采集大量带标签数据。
商汤开源SenseNova-Vision：经典视觉任务并入统一大模型
商汤开源SenseNova-Vision理解生成统一视觉大模型，原生融入目标检测、分割、深度预测、3D重建等经典视觉任务。单模型在结构化理解等四大领域比肩甚至超越专用专家模型，零样本泛化与抗视觉错觉能力强。同步开源5000万条视觉指令语料库。
聆思科技完成近5亿元B轮融资
端侧AI芯片厂商聆思科技完成近5亿元B轮融资，由安徽、合肥多家国资平台战略领投，讯飞创投等机构跟投。资金全部投向新一代端侧认知大模型推理芯片研发。
端侧AI推理正在成为下一个兵家必争之地——从云端到边缘，大模型的落地需要专用芯片支撑。
谷歌Gemma 4突袭开源
谷歌在当日发布Gemma 4开源模型系列，进一步推动开源大模型生态发展。海外方面，Meta生图工具上线仅4天因隐私争议下架。
Physical AI is hitting production readiness faster than expected. July 14 saw two major storylines: physical AI landing beyond expectations, and on-device large models entering an arms race.
Unitree G1 in Nature: The international journal Nature published a landmark paper — a Unitree G1 humanoid robot completed the world’s first remote-controlled live pig laparoscopic cholecystectomy (minimally invasive gallbladder removal). This fills the gap for general-purpose humanoid robots in precision medical scenarios.
BAAI Orca World Model: The Beijing Academy of AI released RoboBrain Orca, a universal world foundation model trained on 125,000 hours of unlabeled real-world video. Without relying on robot action labels, it matches specialized embodied systems on downstream tasks.
SenseTime SenseNova-Vision: SenseTime open-sourced SenseNova-Vision, a unified vision-language model that natively integrates object detection, segmentation, depth prediction, and 3D reconstruction — matching or surpassing specialized expert models across four domains, with 50 million visual instruction samples also released.
Lingsi Technology 500M yuan Series B: Edge AI chipmaker Lingsi closed ~500M yuan Series B led by state-backed investors, with iFlytek Capital participating. Funds go entirely to next-gen edge cognitive LLM inference chips.
Google Gemma 4: Google surprise-released the Gemma 4 open model series, while Meta’s image generation tool was pulled just 4 days after launch due to privacy concerns.
Source: Nature, 163.com, BAAI