ORBIT AI lab revises experiments with human checkpoints

In a laboratory in Nanjing, China, a robot finishes an experiment and AI helps revise the next plan. Science and Technology Daily’s November 13 visit to Lingshu AI Pharma describes ORBIT supporting six or seven research programs. After failures, revised plans require human confirmation before execution. People also retain checkpoints for key choices, irreversible operations and safety.

An earlier public demonstration helps explain the setup. Lingshu presented its multi-agent research-and-validation platform at an industry forum in Nanjing, China on September 12. Xinhua Daily subsequently reported how it combined an internally developed research platform with high-throughput organoid experiments. In drug discovery, a promising computational proposal still has to meet a physical test. Returning those measurements to the planning system gives the next decision something more useful than another prediction alone. Connecting the two sides could also reduce time spent transferring information between separate tools.

The company’s website describes AI designing molecules and predicting their properties, with organoid experiments providing validation. Organoids are miniature tissue models grown for research; lead compounds are promising molecules selected for further improvement. Lingshu claims that the stage from target validation to lead-compound optimization can take six to nine months instead of the traditional two or three years. The claimed acceleration concerns an early research stage. Turning those candidates into usable medicines remains a much longer undertaking.

Think of a cook testing a recipe. If the first batch is too salty, the useful next step is to change the ingredients, rather than faithfully cook the same batch again. A robot resembles the assistant that measures and performs each operation precisely; AI proposes what to adjust, while a person decides which changes are worth trying. Reliable repetition matters, but so does a record explaining why the next experiment was chosen. China’s lights-out laboratory group standard, published and implemented on January 26, provides a framework covering assessment procedures, basic conditions and certificate management.

A visible failure is easier to notice than gradual deterioration. In the interview, chairman Ling Yu identified automatic detection of models quietly getting worse over time as a remaining gap. A system might keep producing results while slowly steering research in a less useful direction. Addressing that problem requires more than a smoother robot arm or a faster experiment. It calls for people who keep comparing outcomes and know when to let the next test proceed, when to pause, and when to revisit the question being asked.

Original source published: Science and Technology Daily, November 13, 2026, 07:39 Beijing time / November 12, 6:39 PM EST; Xinhua Daily, September 13, 2026, 18:26 Beijing time / 6:26 AM EDT.