
刷新权威榜单SOTA!ACE-Ego 解锁“以人为中心”的规模化具身模型训练新范式
Quick Answer
ACE Robotics and CUHK MMLab have launched ACE-Ego, a new human-centric VLA model that achieved SOTA results with a 72.8% success rate on the RoboCasa GR1 TableTop benchmark and 90.62% on RoboTwin 2.0, significantly outperforming competitors like NVIDIA's GR00T and JD's JoyAI-RA.
Quick Take
This model leverages large-scale first-person human video data for effective training, paving the way for advanced robotic applications in retail.
Key Points
- ACE-Ego achieved a 72.8% success rate on RoboCasa GR1 TableTop, setting a new record.
- The model demonstrated 90.62% success in challenging scenarios on the RoboTwin 2.0 benchmark.
- ACE-Ego integrates first-person human video data for enhanced model training efficiency.
- The model significantly outperformed NVIDIA's GR00T and JD's JoyAI-RA in various tasks.
- ACE-Ego enables complex retail operations, including packaging and sorting tasks.
Source Excerpt
近日,大晓机器人联合香港中文大学多媒体实验室(CUHK MMLab)正式发布全新“一脑多型”具身操作VLA模型 ACE-Ego,并向行业开源。 作为“以人为中心”ACE 研发范式在具身模型预训练的核心落地成果,ACE-Ego提出大规模第一视角人类视频与多型机器人数据高效联合预训练的新范式,在两大国际权威具身智能基准上双双领先,并在复杂零售场景中展现出强泛化落地能力,为具身操作模型的规模化演进开辟了全新路径。 在国际公认的人形机器人操作基准 RoboCasa GR1 TableTop 上,ACE-Ego 以72. 8%的平均成功率刷新当前最高纪录,夺得榜首,大幅超越英伟达 GR00T、PI π₀. ₅、京东JoyAI-RA 等主流模型;在高难度双臂操作基准 RoboTwin 2. 0 的强域随机化测试中,ACE-Ego 以90. 62%的成功率展现出远超行业平均水平的环境鲁棒性。 2025年12月,大晓机器人提出“以人为中心(Human-centric)” 的ACE具身研发范式,将人类与物理世界的互动规律作为核心研究起点,构建了一套从“环境式数据采集—开悟世界模型3. 0—具身交互”的全链路技术体系。
源于对 “以人为中心” ACE 范式的深度践行,不同于行业传统 “以机器为中心”、依赖大批量高成本真机遥操作数据的路线,ACE-Ego 将海量低成本的第一视角人类视频转化为可用于模型训练的有效监督信号。 …
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from 雷峰网 AI
See more →
刚刚,GPT 5.6 发布会上,OpenAI 暴露了哪些 Agent 技术路线?
OpenAI's GPT 5.6 integrates ChatGPT and Codex, introducing a for complex task execution, with models Soul, Terra, and Luna for efficient workflow management. The release emphasizes task orchestration, contextual understanding, and robust security measures for enterprise applications.

