5分钟完成机器人纳管、10秒启动跨集群任务，清华大学联合无问芯穹开源具身智能云原生平台RLark

AI 导读

清华大学联合无问芯穹开源具身智能云原生平台RLark，通过统一设备管理、跨集群通信优化和任务级编排，实现机器人纳管5分钟、任务启动10秒，为大规模具身智能实验提供基础设施支持。

Tsinghua University and Wuxian Xinqiong open-source RLark, a cloud-native platform for embodied intelligence, achieving 5-minute device management and 10-second task startup through unified orchestration and cross-cluster communication optimization.

重点速览

• 统一管理机器人、相机等设备资源，实现云边端资源调度复用 Unified management of robots, cameras, and edge devices for cloud-edge resource scheduling
• 任务级跨集群通信优化，提升大包传输吞吐量49% Task-level cross-cluster communication optimization boosts large packet throughput by 49%
• 开源降低基础设施门槛，支持跨地域实验闭环 Open-source lowers infrastructure barriers for cross-regional experiments
• 自研具身运行时将设备转化为可调度资源 Custom embodied runtime converts physical devices into schedulable resources
• 控制面+数据面架构实现任务全链路观测 Control plane + data plane architecture enables end-to-end task observability

一句话：RLark通过云原生架构重构具身智能基础设施，实现复杂任务的高效协同与规模化部署。 / RLark redefines embodied intelligence infrastructure with cloud-native architecture, enabling efficient collaboration and scalable deployment of complex tasks.

推荐理由 该技术突破解决了具身智能规模化部署的核心痛点，为跨地域协同实验提供可复用的基础设施范式。