Faced with the challenge that AI research cannot autonomously carry out offline practical tasks, Huawei's openJiuwen team has unveiled the Jiuwen Symbiosis physical agent. This innovative agent bridges the gap between large-scale models and underlying physical devices, paving the way for the realization of automated scientific research in wet lab environments. At its core, it operates on an Agent loop consisting of "perception—planning—execution—feedback." It constructs a structured representation of the experimental site by amalgamating environmental and device data. Leveraging these structured states, it dynamically coordinates sequences of cross-device subtasks, assigns different execution units to perform physical actions, and propels the next round of decision-making through the verification of execution results. Moreover, the agent is equipped with the capability to retain execution memory and replay execution paths. This enables it to have a real-time understanding of on-site conditions, conduct retrospective analyses of the entire experimental process, and perform clustered analyses of historical failure cases. Simultaneously, it integrates pre-set safety barriers and anomaly recovery mechanisms to guarantee boundary safety and manage faults in a controlled manner during experimental operations. This comprehensive approach forms a complete physical agent system, covering everything from the generation of experimental plans to their offline practical execution.