GPT-6 Astra开进机器人身体！清华联手无问芯穹等开源RPent

AI Digest

清华联合无问芯穹等团队开源RPent，将大模型与专家模型结合，构建具身智能体闭环系统，实现机器人在物理世界完成复杂任务，任务成功率超92.6%。

Tsinghua University and Wuwen Xinqiong open-source RPent, integrating large models with expert models to create a embodied intelligence system for physical-world task execution with over 92.6% success rate.

Key points

• RPent开源连接大模型与机器人，形成感知-决策-执行闭环 RPent connects large models with robots via modular architecture
• 任务成功率92.6%，执行速度优化7倍以上 Achieves 92.6% task success rate with 7x speed optimization
• 模块化架构支持多机器人/仿真环境接入 Supports multi-robot/simulation environment integration
• 记忆系统实现经验沉淀与持续学习 Memory system enables experience accumulation
• Flash Mode提升稳定任务执行效率 Flash Mode enhances stable task execution efficiency

Takeaway: RPent为具身智能提供开放基础设施，推动大模型真正落地物理世界。 / RPent provides open infrastructure for embodied intelligence, enabling large models to operate in physical worlds.

Why it matters 展示大模型与机器人协同的创新框架，为实际应用提供可扩展技术路径。