MagicAtom's EchoPolicy-0 and MagicBrain Frameworks Achieve Top Performance in MolmoSpaces Benchmark
魔法原子具身 Agent 框架拿下 MolmoSpaces 全球第一 让 VLA 从“演示可用”逼近“产线可用”
MagicAtom's EchoPolicy-0 framework achieved a 75.4% average success rate in MolmoSpaces Combined benchmark, ranking first globally, while MagicBrain achieved 72.4% with a 94.4% grasp success rate. Both frameworks address the challenge of aligning language instructions with visual perception in embodied AI, with EchoPolicy-0 focusing on trajectory selection and MagicBrain using visual guidance for task alignment.
The paper presents two VLA-based frameworks, EchoPolicy-0 and MagicBrain, achieving high success rates in MolmoSpaces benchmark tasks. EchoPolicy-0 demonstrates a unified VLM Agent framework with trajectory selection and validation, while MagicBrain uses visual guidance to improve task alignment. Both approaches highlight the importance of visual understanding in VLA systems.
Source: Chinese robotics — Robot learning research · 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%
View original evidence
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 S3Grasp-and-place success rate
94.4%
View original evidence
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
- paper: 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.
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
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
EchoPolicy-0 VLM 框架多候选轨迹选择原理 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
EchoPolicy-0 针对具身智能模型落地场景，回答了一个问题：模型提出来的动作，怎么保证选的是对的那一个。
Open source S21
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
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
视觉-语言-动作（VLA）是当前具身智能领域备受关注的一类模型架构，其语言跟随能力是行业内一个由来已久的共性问题。
Open source S24
Source：Chinese robotics — Robot learning research · leiphone.com