Genie Master: A Framework for Embodied Agents to Evolve in Real-World Scenarios
Overview
Genie Master enables robots to self-improve in real-world settings. Xingyuan's data engine enhances training data by 40x, cutting costs to 5 yuan/hour. Both tools aim to advance embodied AI in robotics applications. Latest: Genie Master framework released to support real-world robot evolution. Xingyuan's data engine cuts training costs to 5 yuan/hour with 40x data expansion.
Generated from attributed reports · Updated 23 minutes ago
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · effective training data expansion: 40 other · Basis not reported · Differing source assertions
“行元发布机器人跨本体数据引擎Chorus：“不挑设备”，可实现约40倍训练数据扩容”
Exact source · revision 2Source owner not reported
Reported quantity · effective training data expansion: 40 other · Basis not reported · Differing source assertions
“在当前测试配置下，Chorus 通过 16 种机器人本体适配，将每小时有效机器人轨迹的处理成本降至 15 元；结合 3 倍指令增广，可实现约 40 倍的有效训练数据扩容，单位训练样本的等效成本仅需 5 元/小时。”
Exact source · revision 2Source owner not reported
Reported quantity · cost per hour of effective robot trajectories: 15 other · Basis not reported
“在当前测试配置下，Chorus 通过 16 种机器人本体适配，将每小时有效机器人轨迹的处理成本降至 15 元；结合 3 倍指令增广，可实现约 40 倍的有效训练数据扩容，单位训练样本的等效成本仅需 5 元/小时。”
Exact source · revision 2Source owner not reported
Reported quantity · equivalent cost per training sample: 5 other · Basis not reported
“在当前测试配置下，Chorus 通过 16 种机器人本体适配，将每小时有效机器人轨迹的处理成本降至 15 元；结合 3 倍指令增广，可实现约 40 倍的有效训练数据扩容，单位训练样本的等效成本仅需 5 元/小时。”
Exact source · revision 2Source owner not reported
Reported quantity · number of robot embodiments supported: 16 other · Basis not reported
“在当前测试配置下，Chorus 通过 16 种机器人本体适配，将每小时有效机器人轨迹的处理成本降至 15 元；结合 3 倍指令增广，可实现约 40 倍的有效训练数据扩容，单位训练样本的等效成本仅需 5 元/小时。”
Exact source · revision 2Source owner not reported
Reported quantity · video frames: 17000000 other · Basis not reported
“为支撑大脑训练，大晓开源了ACE-Data-0数据集，包含1700万帧视频、200个任务类别、50名参与者、2个真实家庭场景与7.5万个交互片段。”
Exact source · revision 2Source owner not reported
Reported quantity · task categories: 200 other · Basis not reported
“为支撑大脑训练，大晓开源了ACE-Data-0数据集，包含1700万帧视频、200个任务类别、50名参与者、2个真实家庭场景与7.5万个交互片段。”
Exact source · revision 2Source owner not reported
Reported quantity · participants: 50 other · Basis not reported
“为支撑大脑训练，大晓开源了ACE-Data-0数据集，包含1700万帧视频、200个任务类别、50名参与者、2个真实家庭场景与7.5万个交互片段。”
Exact source · revision 2Source owner not reported
Reported quantity · real-world scenes: 2 other · Basis not reported
“为支撑大脑训练，大晓开源了ACE-Data-0数据集，包含1700万帧视频、200个任务类别、50名参与者、2个真实家庭场景与7.5万个交互片段。”
Exact source · revision 2Source owner not reported
Reported quantity · interaction fragments: 75000 other · Basis not reported
“为支撑大脑训练，大晓开源了ACE-Data-0数据集，包含1700万帧视频、200个任务类别、50名参与者、2个真实家庭场景与7.5万个交互片段。”
Exact source · revision 2Source owner not reported
Reported quantity · sim2real transfer: 1 other · Basis not reported
“据其官网与论文介绍，GraspVLA模型基于名为SynGrasp-1B的十亿帧机器人抓取合成数据完成预训练，把自回归感知与基于流匹配的动作生成整合进统一过程，可直接由仿真到现实（Sim2Real）零样本迁移至未见场景；公司的数据基建"银河星坊（AstraSynth）"据其公开介绍可大规模量产合成操作数据。”
Exact source · revision 2Source owner not reported
Reported quantity · human first-person video hours: 10000 hours · Basis not reported
“为了训练这一模型，研究团队构建了 Xp-Data，整合约 1 万小时人类第一视角视频和 3000 小时机器人视频 ，统一相机轨迹、运动骨架和任务描述等多模态标注。”
Exact source · revision 2Source owner not reported
Reported quantity · robot video hours: 3000 hours · Basis not reported
“为了训练这一模型，研究团队构建了 Xp-Data，整合约 1 万小时人类第一视角视频和 3000 小时机器人视频 ，统一相机轨迹、运动骨架和任务描述等多模态标注。”
Exact source · revision 2Source owner not reported
Reported quantity · PSNR score: 21.98 other · Basis not reported
“在覆盖八种本体的跨本体评测中，同一组 Ropedia W0 权重无需针对目标评测数据额外训练，就取得了 21.98 dB 的 PSNR，高于使用相同机器人混合数据适配的 Cosmos3-Nano（19.02 dB），展现出跨本体的世界预测能力。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 56.82 percent · Basis not reported · Differing source assertions
“在 LIBERO-Plus 基准上，Ropedia W0 初始化版本的成功率由 50.06% 提升至 56.82% ，并在相机、光照、背景等七类环境扰动下均取得改善。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 50.06 percent · Basis not reported · Differing source assertions
“在 LIBERO-Plus 基准上，Ropedia W0 初始化版本的成功率由 50.06% 提升至 56.82% ，并在相机、光照、背景等七类环境扰动下均取得改善。”
Exact source · revision 2Source owner not reported
Evidence dependency changes (4)
- 2026-10-30T17:41:50.036Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T17:41:21.796Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T17:38:03.348Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T17:36:35.946Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
Developments
- 2026-10-27 11:35 UTC · 1 reportsGenie Master: A Framework for Embodied Agents to Evolve in Real-World ScenariosRobotics — Chinese web discovery：Genie Master: A Framework for Embodied Agents to Evolve in Real-World Scenarios
- 2026-10-27 04:39 UTC · 1 reportsXingyuan Releases Cross-Embodiment Data Engine Chorus: Supports Unified ProcessiChinese robotics — Research news discovery：Xingyuan Releases Cross-Embodiment Data Engine Chorus: Supports Unified Processing of Data from Multiple Devices, Achieving Approximately 40x Training Data Expansion
Report timeline
Follow attributed reports and material updates.
- Robotics — Chinese web discoverySignalGenie Master: A Framework for Embodied Agents to Evolve in Real-World Scenarios
Intelligence Yuan has released the Genie Master framework, which allows robots to continuously self-improve in real-world settings without human intervention. The framework enables robots to identify failures, adjust strategies, and accumulate experience, supporting tasks like grocery restocking and food packaging in real stores.
- Chinese robotics — Dataset and collection discoverySignalHow Human Data Becomes Real Robot Capabilities: Ropedia Releases Six Research Findings
During Ropedia Super Week, the Ropedia team released six research findings exploring how human experience can be transformed into training data for robot learning. The research covers cross-robot world prediction, object trajectory recording, first-person reconstruction, programming agent enhancement, bimanual collaboration strategies, and code-driven robot policies. These results demonstrate the diverse applications of large-scale human experience in robot learning and provide new directions for reducing the need for robot-specific data.
- Chinese robotics — Research news discoverySignalXingyuan Releases Cross-Embodiment Data Engine Chorus: Supports Unified Processing of Data from Multiple Devices, Achieving Approximately 40x Training Data Expansion
Xingyuan Intelligent released the Cross-Embodiment Data Engine Chorus, which supports converting human operation videos collected from various devices into pre-trained data suitable for different robot embodiments. Testing shows it can achieve approximately 40x effective training data expansion, reducing the equivalent cost per training sample to 5 yuan per hour.
- Chinese robotics — Robot learning researchSignalFinding the Thinker Behind the Machine: Four Categories of Embodied Brain Companies in 2026
This article categorizes companies working on embodied intelligence into four types: independent brain companies, integrated hardware-software manufacturers, tech giants and large model firms, and compute and open platform enterprises. It highlights ACE ROBOTICS as a leading independent brain company with its Kairos 3.1 world model achieving top results in multiple benchmarks. The piece also discusses challenges such as data scarcity and simulation-to-reality gaps, and provides a framework for selecting the right company based on technical, open-source, and application criteria.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).