Humanoid Robots Start Working in National-Level Labs to Automate Experiments

The robots are designed to replicate the operating logic of human experimenters. Unlike typical six-axis industrial robots, Monte2 can execute long-sequence, flexible tasks on the lab bench, coordinating multiple instruments. A lab director said the introduction of humanoid robots alleviates bottlenecks caused by manual operation and improves consistency. The lab plans to expand the robot fleet to around 100 units by the end of 2027, controlled by AI as a coordinated cluster, a vision the director called an "industrial revolution in experimental science."

Before this deployment, Monte2 had already been used at BGI Manufacturing, the genomics company, gaining practical experience. The move reflects a broader global trend toward autonomous laboratories. In 2024, a University of Liverpool team demonstrated a robot that ran experiments continuously for eight days, choosing reactions and using lab instruments autonomously, with results published in Nature. In the United States, projects like Berkeley's A-Lab synthesized 41 new materials in 17 days, while researchers at North Carolina State University sped up data collection by tenfold.

Large language models are also adding intelligence into labs. Systems such as CMU's Coscientist, Google DeepMind's Co-Scientist, and Sakana AI's AI Scientist are exploring how AI can understand literature, design experiments, and propose hypotheses. In China, the national lab and Yuanluo Technology are pushing AI for Science from computation into physical execution, aiming for what they call "Lab 3.0."

Lab 3.0 is described as a shift from human-centric and equipment-centric labs to self-driven experiments with embodied intelligence. The top layer uses large models to design experimental methods, while the bottom layer uses robots to execute actions, closing the loop between operation and analysis. Yuanluo leverages its proprietary OPN "object-centric physical native model" to help robots understand the physical environment, recognize reagents, cells, and instruments, and grasp interactions between objects, rather than just following preset programs.

The robots have demonstrated continuous autonomous operation for several hours, maintaining precision at sub-millimeter level while handling tasks like cell passaging and cytotoxicity testing. They integrate visual, force, and tactile feedback to adjust motions in real time, adapting to variations in liquid levels and tube angles.

In biological experiments, such as cell toxicity assays, the robots handle dozens of repetitive steps, reducing the burden on researchers and avoiding human errors caused by fatigue or experience differences. Yuanluo says the robots can complete over 40 refined biological operations stably.

Milan Abolhasani, a professor at North Carolina State University, noted that autonomous labs will act as collaborators to shorten time and costs but will not replace the unique expertise and creativity of human researchers. The shift may free scientists from repetitive work, allowing them to focus on areas requiring human judgment.

To promote Lab 3.0, Yuanluo has launched the "Origin Program" to recruit 100 partners from universities, research institutes, and biopharma companies to validate embodied AI in applications like biomedicine, materials science, and chemical analysis.