PulseAugur
EN
LIVE 23:44:25
中文(ZH) 星海图新一代具身基础模型G0.5发布,全面提升零样本泛化能力,让机器人边思考边行动

OpenGalaxea ships G0.5 embodied model for robots that think while acting

Open-source AI lab OpenGalaxea has released G0.5, a new embodied foundation model designed to enable robots to think while acting. This model integrates reasoning and action into a single architecture, allowing robots to perform tasks with zero-shot generalization. G0.5 utilizes a unified action decoder and a native action Chain-of-Thought mechanism, enabling it to decompose long-range tasks and adapt its actions based on natural language prompts. The model also incorporates a spatio-temporal attention module for enhanced contextual awareness, demonstrating state-of-the-art performance on various benchmarks. AI

IMPACT Enables robots to perform complex tasks with natural language commands, potentially accelerating adoption of embodied AI in various industries.

RANK_REASON New embodied foundation model release from an AI lab. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on 雷峰网 (Leiphone) →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

OpenGalaxea ships G0.5 embodied model for robots that think while acting

COVERAGE [1]

  1. 雷峰网 (Leiphone) TIER_1 中文(ZH) ·

    Xinghai Tu releases new generation embodied foundation model G0.5, comprehensively improving zero-shot generalization capabilities, enabling robots to think and act simultaneously

    <p>过去,机器人听到“帮我把毛巾放进洗手池”——它只会机械地执行一串预设动作。现在,我们让G0.5做到了:先想——毛巾在哪?洗手池在哪?应该先拿起毛巾,再移动过去,最后放下。边思考,边行动。</p><p>这一能力依托于我们最新的 VLA(视觉-语言-行动)模型架构设计:让同一个模型、同一套权重,同时完成推理与行动,让机器人拥有“行动中的思考能力”。</p><h4><span>言出法随:零样本泛化至新物体、新环境、新指令</span></h4><p>G0.5 在 R1 Lite 机器人上,零样本(未经任何该场景的微调),仅凭自然语言指令直接驱动机器人完成…