Magic Atomic's Embodied Agent Framework Secures First Place in MolmoSpaces Global Rankings
魔法原子具身Agent框架拿下MolmoSpaces全球第一
The EchoPolicy-0 and MagicBrain frameworks developed by Magic Atomic achieved first and fourth place, respectively, in the MolmoSpaces global evaluation. EchoPolicy-0 achieves multi-task shared reasoning and action through a unified VLM Agent framework, while MagicBrain employs context-based visual guidance to enhance target recognition capabilities. Both have made significant progress in the field of embodied intelligence, advancing the transition from demonstration usability to production usability.
Source: QbitAI — Robotics · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
94.4%
Reported success rate
94.4%
Grasp-and-place success rate
94.4%
View original evidence
抓取（Pick-v1）、抓取放置（Pick & Place-v1）两项任务均拿下了所在任务榜单的全球第一，抓取任务成功率更是高达94.4%
Open source S23
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
< img id="wx_img" src="https://www.qbitai.com/wp-content/uploads/imgs/qbitai-logo-1.png" width="400" height="400"> 2026-10-13 19:05:39 来源： 量子位 让VLA从“演示可用”逼近“产线可用” 近日，在由艾伦人工智能研究所（AI2）发起的全球权威具身智能零样本评测基准MolmoSpaces最新榜单上，魔法原子旗下面向通用机器人操作的VLM Agent框架EchoPolicy-0以75.4%的机器人操作平均成功率位列MolmoSpaces Combined榜单全球第一，覆盖全部4个v1任务。
Open source S2
EchoPolicy-0框架在MolmoSpaces的四项操作任务上完成了3915局全量评测，取得了75.4% 的四项任务等权平均成功率，截至2026年10月13日，位列MolmoSpaces Combined榜单球第一。
Open source S19
自研大脑：MagicBrain让模型能“看见” 从“演示可用”逼近“产线可用” 具身智能Agent框架MagicBrain来自魔法原子MagicLab Studio ，在MolmoSpaces的抓取、抓取放置、打开与关闭四项操作任务上， MagicBrain完成了3915局全量评测，四项v1任务平均成功率全球第四。
Open source S22
同一榜单、同一分组下，魔法原子自研具身智能Agent框架MagicBrain凭借3.4B参数规模，以72.4%的机器人操作平均成功率位列MolmoSpaces Combined榜单全球第四，其抓取成功率达到94.4%，远超行业平均水平。
Open source S3
其中，抓取（Pick-v1）、抓取放置（Pick & Place-v1）两项任务均拿下了所在任务榜单的全球第一，抓取任务成功率更是高达94.4%。
Open source S23
Source：QbitAI — Robotics · qbitai.com