Skip to content
Robotics — Chinese web discovery·· 4 days agoSignalEditorial score75

AGIBOT Unveils GO-3 Preview Embodied AI Model, Tops Three Major Benchmarks

智元机器人发布GO-3 Preview模型,登顶三大具身智能评测榜单

Summary

AGIBOT recently released the GO-3 Preview embodied AI model, which achieved first place in three major benchmarks: RoboDojo, RoboTwin, and RoboColiseum. Built on AGIBOT's proprietary WRAM architecture, the model supports full-lifecycle learning spanning pre-training, post-training, and deployment-time adaptation. It is planned to be fully open-sourced by the end of 2026.

Source: Robotics — Chinese web discovery · 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.

  • paper: not_reported
  • code: not_reported
  • weights: not_reported
Source excerpts and review record

Automatically extracted; no manual editorial approval recorded.

人类第一人称视频提供了丰富的物体交互与任务执行示例,而机器人轨迹数据则建立了观察、动作与其物理结果之间的联系。

Open source S18

GO-3 Preview从人类第一人称视频与机器人轨迹数据中学习,将物体交互与任务执行同动作及物理结果相连接。

Open source S19

抓取小物体、组装零件等任务,要求机器人具备预判环境变化、控制接触力度以及执行精确物理动作的能力。

Open source S7

该任务要求机器人识别单个零件、精确抓取、对齐位置并正确插入。

Open source S21

在零样本场景中,该模型无需针对特定任务在实体机器人上进行监督微调,即可将先前学习的能力应用于新任务。

Open source S20

智元机器人(AGIBOT)近日发布了新一代具身基础模型GO-3 Preview,该模型支持跨越预训练、后训练及部署后学习全生命周期的大规模学习。

Open source S2

该模型已在三大主要具身智能评测基准——RoboDojo、RoboTwin和RoboColiseum中取得第一名,并计划于2026年底前完全开源。

Open source S3

GO-3 Preview基于智元机器人自主研发的具身原生架构WRAM(World Reasoning Action Model,世界推理行动模型)构建,将任务理解、未来状态预测与动作生成统一起来,使机器人能够在不同任务间实现泛化,并完成精确的操作动作。

Open source S4

后训练扩展:通过物理与虚拟交互实现提升 仅凭通用能力,机器人无法可靠地完成每一项复杂任务。

Open source S23

Source:Robotics — Chinese web discovery · msn.cn

Timezone · UTC

Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).