Shanghai AI Laboratory is accelerating the construction of a distributed autonomous laboratory network, providing underlying support for the Intern·Duanyan scientific discovery platform, enabling cross-regional connectivity, on-demand scheduling, and continuous evolution of experimental capabilities.
Recently, the laboratory operating system Intern·Tianshu (Sage·LabOS) was launched, aiming to better organize collaborative wet experiments across laboratories. It uses the experimental equipment access and collaboration protocol Intern SCP-Lab as a unified interface, and leverages the Intern Physical World Model W0 to drive robots to complete experimental operations, opening up the complete chain from autonomous execution of heterogeneous devices to the return of experimental results to the research AI agent.
Intern·Tianshu has been validated in the lipid nanoparticle (LNP) formulation optimization task guided by the 'Yuansheng' platform at the Lingang Laboratory, supporting five rounds of dry-wet closed-loop iterations, automatically completing the synthesis and validation of a total of 335 formulations, helping the encapsulation rate jump from 67% to 98%, with in vitro transfection levels improving round by round, fully demonstrating its ability to support continuous operation of autonomous laboratories and multi-round experimental iterations.
With the successive launch of a series of research advances such as Intern W0, Intern SCP-Lab, and Intern·Tianshu, the distributed autonomous laboratory network of Shanghai AI Laboratory has begun to take shape. Research institutes, universities, and industry partners are invited to connect to the network, jointly build a scientific intelligence ecosystem, and explore new paradigms of AI-driven scientific discovery.
Diagram of the distributed autonomous laboratory network: Building four core capabilities to create a self-evolving hub for autonomous laboratories. Autonomous laboratories are moving from single-point automation to full-process autonomous operation.
While research AI agents, robots, and automated instruments continue to iterate, problems such as inconsistent interfaces of heterogeneous devices, difficulty in software-hardware collaboration, reliance on manual sample transfer and data return, and high cost of experimental integration are becoming increasingly prominent.
The industry urgently needs an operating hub that is compatible with diverse equipment, can be quickly deployed, and supports continuous evolution. Intern·Tianshu covers the entire experimental execution chain, relying on four core capabilities: capability discovery, task orchestration, execution scheduling, and data return, to connect device interfaces, robot actions, task status, and experimental data, supporting collaborative operation, rapid deployment, and iterative upgrade of autonomous laboratories.
Intern·Tianshu supports collaborative operation, rapid deployment, and iterative upgrade of autonomous laboratories. Capability discovery: Unify device capability descriptions, compatible with multiple control methods. Relying on Intern SCP-Lab to standardize device capability definitions, call logic, and feedback information, Intern·Tianshu can automatically discover available laboratory equipment; paired with the ADK access suite to encapsulate driver interfaces, it supports multiple control modes such as API direct connection and GUI intelligent agent adaptation, allowing existing instruments to be quickly connected, and once connected, capabilities can be reused across processes.
Task orchestration: Link instruments and robots, turning experimental plans into coordinated actions. For complex experimental scenarios involving multiple instruments and multiple transfers, Intern·Tianshu parses the experimental plan, matches instrument functions with robot skills, and connects steps such as sample preparation, transfer, and characterization.
Robots rely on Intern W0, comprehensively using perception, navigation, and motion planning to complete on-site operations such as sample pickup and transfer. Execution scheduling: Coordinate step dependencies and resource status to ensure safe and orderly experimental operation.
Experimental steps have sequential dependencies, and robots and instruments each have their own operating status and resource occupancy. Intern·Tianshu judges in real time whether previous steps are completed, device occupancy, and resource status, then performs resource matching and task scheduling, continuously tracks execution status, and promptly handles anomalies.
Device access and experimental execution are subject to security policies throughout. Researchers can view progress uniformly without repeatedly checking across devices. Data return: Close the experimental loop, driving continuous optimization of plans.