The robots can analyze and improve on experimental methods themselves
In a Tokyo laboratory, ten two-armed robots have begun conducting biological experiments with a degree of independence that was, until recently, the province of science fiction. The Institute of Science Tokyo's automated facility represents a quiet but consequential shift in how humanity pursues knowledge — not by replacing the scientist's mind, but by liberating it from the tyranny of repetition. What is unfolding here is less a technological spectacle than a philosophical reorientation: the question is no longer only what we can discover, but how we might discover it faster, and together, at a scale the individual researcher could never achieve alone.
- Ten AI-guided robots are already conducting complex biological experiments autonomously in Tokyo, including an 111-day stem cell optimization trial that would have consumed months of human effort.
- The ambition stretches far beyond ten machines — researchers envision a factory-scale facility of thousands of robots available to scientists worldwide by 2040 to 2050, a vision that would restructure global research infrastructure.
- The gap between today's working prototype and that future remains wide: humans still prepare reagents, troubleshoot failures, refill consumables mid-experiment, and intervene when a robot drops a vial.
- Scientists outside Tokyo are watching closely, comparing the potential to CERN — a shared international resource — while cautioning that true laboratory autonomy is still years of advanced programming away.
- The most significant tension is not mechanical but conceptual: these robots do not merely execute protocols, they analyze and improve them, blurring the line between tool and collaborator.
Inside a Tokyo laboratory, ten two-armed robots pipette liquids, seed cell cultures, and operate instruments with mechanical precision. Opened in April at the Institute of Science Tokyo's Robotics Innovation Center, the facility represents years of pursuit toward experiments that largely run themselves. Researcher Genki Kanda and his team built something that works — but their real ambition is far larger: a factory-scale operation of thousands of robots, accessible to scientists around the world by 2040 or 2050.
The robots carry more than mechanical capability. AI software allows them to make decisions about experiments themselves. One program tested 144 experimental conditions over 111 days to find the optimal environment for growing human stem cells. In another trial, robots imaged cells, predicted their development, and determined the precise moment to harvest them — maintaining cultures for eight consecutive days while the research team was away on holiday.
Observers elsewhere see the significance. Vanderbilt materials scientist Yan Zeng notes that two-armed robots represent a genuine leap beyond older single-arm systems, enabling tasks of real complexity. She imagines the Tokyo facility eventually functioning like CERN — open international infrastructure that accelerates discovery across disciplines. University of Liverpool chemist Andrew Cooper adds a sharper point: these machines are not simply automating existing work, they are analyzing and improving experimental methods, learning from results rather than merely executing protocols.
Yet the distance between today and Kanda's vision remains real. Reagents are still prepared by hand. Humans troubleshoot failures, refill consumables mid-experiment, and intervene when equipment drops a vial. Kanda's team is developing software to push past these limits — enabling robots to adapt when errors occur and adjust protocols based on available resources. Cooper cautions that fully autonomous laboratories are still years away, the integration of AI with physical systems remaining largely in the proof-of-concept phase. In Tokyo, though, the robots are working, learning, and returning something precious to the researchers who built them: time to think.
In a Tokyo laboratory, ten robots with two articulated arms move through the motions of science. They pipette liquids with mechanical precision. They seed cell cultures onto petri dishes. They operate instruments that would take a human researcher hours to manage alone. This is the Robotics Innovation Center at the Institute of Science Tokyo, which opened its automated laboratory in April, and it represents something researchers have been chasing for years: the possibility of experiments that largely run themselves.
Genki Kanda, an automation researcher at the center, oversees the facility. He and his team have built something that works—ten robots that can handle the repetitive, time-consuming labor that fills a biologist's day. But they are thinking much larger. The real ambition is to construct what Kanda calls a "factory-scale" operation: thousands of robots, all capable of independent work, all available to scientists around the world by 2040 or 2050. It is a vision that would fundamentally reshape how research gets done.
The robots do more than follow instructions. They contain artificial intelligence software that allows them to make decisions about the experiments themselves. An AI program developed by Kanda's team tested 144 different experimental conditions over 111 days to determine the optimal environment for growing human stem cells—work that would have consumed months of human labor. In another trial, the robots imaged cells as they grew, predicted their development over time, and determined the precise moment to harvest them. For eight consecutive days, while Kanda's team was away on holiday, the robots maintained those cell cultures without intervention.
Yan Zeng, a materials scientist at Vanderbilt University, sees the potential. She notes that two-armed robots represent a genuine leap beyond the one-armed systems that have existed in some labs for a decade. With two arms, the machines can perform tasks of genuine complexity and sophistication. Zeng is curious whether the Tokyo team will actually reach their 2040 target, but she imagines a future where such a facility operates like CERN or other major international research infrastructure—open to scientists globally, a shared resource that accelerates discovery across fields.
Andrew Cooper, a chemist at the University of Liverpool, emphasizes that the robots are not merely automating existing work. They are analyzing experimental methods and improving them. The AI does not just execute a protocol; it learns from results and adapts. This distinction matters. It suggests a future where machines do not simply replace human hands but augment human thinking.
Yet the gap between what exists now and what Kanda envisions remains substantial. His team still prepares reagents and materials by hand. They troubleshoot when something goes wrong. They clean up afterward. When the robots run low on consumables mid-experiment, humans refill them. When equipment fails, humans fix it. When the robots make mistakes—dropping a vial, for instance—a person must intervene. Kanda's group is working to push these boundaries, developing software that would allow robots to act as autonomous scientists, making decisions when errors occur and adapting protocols based on what equipment and resources are actually available.
Cooper cautions that fully autonomous laboratories remain years away. Integrating artificial intelligence into physical robots is genuinely difficult work. It demands advanced programming skills and remains largely in the proof-of-concept phase. The dream of a lab that runs without human presence is not yet real. But in Tokyo, the robots are working. They are learning. And they are buying researchers time to think.
Citas Notables
The lab's ultimate goal is to create a factory-scale facility with thousands of robots that could be used by local and international scientists by 2040 or 2050— Genki Kanda, automation researcher at the Robotics Innovation Center
Two-armed robots can do more complicated and sophisticated tasks than one-armed systems— Yan Zeng, materials scientist at Vanderbilt University
La Conversación del Hearth Otra perspectiva de la historia
What makes two arms so much better than one?
One arm can move samples around, position things, operate basic equipment. Two arms can do what humans do—hold something steady with one hand while manipulating it with the other. You can pour while measuring. You can hold a plate while inoculating it. The complexity jumps.
So the AI isn't writing the experiments from scratch?
No. It's working within protocols humans design. But it's making real decisions—testing variations, analyzing results, deciding what to try next. In the stem cell work, it ran through 144 conditions in 111 days. A human would take months.
Why do they still need people at all?
Because the robots don't understand context the way humans do. If something breaks, if a reagent runs out unexpectedly, if the equipment isn't quite what the protocol assumed—those moments require judgment. The robots can follow a plan beautifully. They struggle when the plan needs to bend.
Is this about replacing scientists?
It's about freeing them. Right now, researchers spend enormous time on repetitive work. The robots handle that. The humans design better experiments, interpret stranger results, ask questions the robots wouldn't think to ask.
When do we actually get the thousands-of-robots vision?
That's the honest question. Kanda says 2040 or 2050. Cooper says full autonomy is still years away. The engineering is hard. But the direction is clear.